Arxiv:2009.14053V3 [Math.GR] 24 Jun 2021 1.1

arXiv:2009.14053v3 [math.GR] 24 Jun 2021 1.1. spaces. hierarchical shortcut strongly curvature: and spaces, nonpositive form injective w exhibiting article, various this spaces satisfying In spaces of fruitful. spaces on especially metric been acting has of groups conditions types of the study via the groups studying includes This oi,heacial uscne,buddpacking. bounded quasiconvex, hierarchically bolic, obe to the describing postpone We 3.1. u [ANS19]. Section establishing prov groups and as cubical [BHS17c] such many groups successes, for of class number mapping a cubical exampl for had known jecture interesting has all theory of and The [BHMS20] body groups others. growing class mapping a spaces of These is tients [BHS17b]. there in and Sisto behaviour, and Hagen, Behrstock, by duced OREIJCIIY IRRHCLHPROIIY AND HYPERBOLICITY, HIERARCHICAL INJECTIVITY, COARSE h rto u he lse sta of that is classes three our of first The gr of study the is theory group geometric of theme principal A e od n phrases. and words Key 2020 Date h etcasw osdri hto oreyijciespac injective coarsely of that is consider we class next The h setting. The oreyinjective coarsely ue2,2021. 25, June : ahmtc ujc Classification. Subject Mathematics rtdaeinsbrusaeudsotd ln h a es we group way hyperbolic the hierarchically and Along of subgroups, subgroups undistorted. quasiconvex finite are of subgroups abelian classes erated conjugacy have many semihyperbolic, finitely are have they that including groups, bolic short strongly are groups injective coarsely and injective clas stron mapping is groups—including geometry hyperbolic s hierarchically bounded We uniformly of space. space metric hyperbolic m jective original hierarchically the the to m of quasi-isometric new is automorphisms a metric new admits st The space and jective. hyperbolic spaces, hierarchically any injective that coarsely show spaces, hyperbolic cally Abstract. sn hs eut,w euesvrlipratpropertie important several deduce we results, these Using HMSHETL IAHD,ADHRYPETYT HARRY AND HODA, NIMA HAETTEL, THOMAS erlt he lse fnnoiieycre ercspac metric curved nonpositively of classes three relate We fteei constant a is there if irrhclyhproi,casl netv,strongly injective, coarsely hyperbolic, hierarchically SEMIHYPERBOLICITY 1. Introduction 06,2F7 51F30. 20F67, 20F65, irrhclyhproi spaces hyperbolic hierarchically 1 δ uhta o n family any for that such r ocre ihtreclasses three with concerned are e ruso ufcsaecoarsely surfaces—are of groups s cut. aebuddpacking. bounded have s ogysotu pcs We spaces. shortcut rongly ovbecnuayproblem, conjugacy solvable ti n speevdunder preserved is and etric l hrct Consequently, shortcut. gly yhproi pcs coarsely spaces, hyperbolic ly o htaycasl in- coarsely any that how fheacial hyper- hierarchically of s ti hti oreyin- coarsely is that etric o hthierarchically that how htterfiieygen- finitely their that iomepnnilgrowth exponential niform fnnoiiecurvature nonpositive of s hyato.I hsvein, this In on. act they s ercsaei said is space metric A es. n absqaiascon- quasiflats Farb’s ing s nldn ayquo- many including es, irrh tutr until structure hierarchy rus[S0,amongst [HS20], groups usa ercspaces. metric as oups xii hyperbolic-like exhibit s hierarchi- es: hc eeintro- were which , hrct semihyper- shortcut, t B p x i r , i q : i P I u of balls with dpxi,xjqď ri ` rj for all i, j P I, the δ–neighbourhoods of those balls have nonempty total intersection. This property was first considered by Chepoi and Estellon in [CE07]. As the term suggests, this notion is closely related to that of injective metric spaces. A metric space is injective (also called hyperconvex) if for any family tBpxi, riq : i P Iu of balls with dpxi,xjqď ri ` rj for all i, j P I, the balls have nonempty total intersection. In other words, it amounts to taking δ “ 0 in the definition of coarse injectivity. (There are multiple equivalent ways to define injectivity of a metric space, by a theorem of Aronszajn and Panitchpakdi [AP56].) A construction of Isbell [Isb64], which was later rediscovered by Dress [Dre84] and by Chrobak and Larmore [CL94], shows that every metric space has an essentially unique injective hull. More precisely, the injective hull of a metric space X is an injective metric space EpXq, together with an isometric embedding e : X Ñ EpXq, such that no injective proper subspace of EpXq contains epXq. For convenience, we will often identify X with its image epXq. A nice description of the construction of the injective hull is given by Lang in [Lan13, §3]. The classes of coarsely injective spaces and injective spaces are tied together by the following useful fact, the proof of which is identical to that of [CCG`20, Proposition 3.12]. A subset Y of a metric space X is coarsely dense if there exists r such that every x P X is r–close to some y P Y . Proposition 1.1. A metric space is coarsely injective if and only if it is coarsely dense in its injective hull. Moreover, if a group acts properly and coboundedly on a coarsely injective space, then it acts properly and coboundedly on the injective hull of that space (see Lemma 3.10). Here and throughout the paper, a group G is said to act properly on a metric space X if tg P G : gB X B ‰ Hu is finite for every metric ball B of X. This is sometimes referred to as a metrically proper action. Injective metric spaces satisfy a number of properties reminiscent of nonpositive curvature, and in particular of CAT(0) spaces. For instance, they admit a conical geodesic bicombing [Lan13], and proper injective spaces of finite combinatorial dimension have a canonical convex such bicombing [DL15]. Also, every bounded group action on an injective metric space has a fixed point, and the fixed point set is itself injective [Lan13]. These properties are what allow us to draw our conclusions for hierarchically hyperbolic groups. Although it will not be needed here, it is interesting to note that injective spaces are also complete [AP56] and contractible [Isb64]. The strong shortcut property was introduced by the second named author for graphs [Hod18] and then generalized to roughly geodesic metric spaces [Hod20b]. A Riemannian circle S is the circle S1 endowed with a geodesic metric of some length |S|. A roughly geodesic metric space pX,σq is strongly shortcut if there exists K ą 1 such that for any C ą 0 there is a bound on the lengths |S| of pK,Cq–quasi-isometric embeddings S Ñ X of Riemannian circles S in pX,σq. A group is strongly shortcut if it acts properly and coboundedly on a strongly shortcut metric space. Many spaces and graphs of interest in geometric group theory and metric graph theory are strongly shortcut, including Gromov-hyperbolic spaces, 1–skeletons of finite-dimensional CATp0q cube complexes, Cayley graphs of Coxeter groups, and asymptotically CATp0q spaces. Despite being such 2 a unifying notion, it remains possible to draw conclusions about strongly shortcut groups, including that they are finitely presented and have polynomial isoperimetric function, and so have decidable word problem. 1.2. Comparison of the classes. Our main result is the definition of a new metric on hierarchically hyperbolic spaces, and more generally on coarse median spaces satisfying a nice approximation property of median intervals by CAT(0) cube complexes. Our construction is directly inspired by work of Bowditch (see [Bow20]), in which he constructs an injective metric on any finite rank metric median space. Indeed, if one endows a finite-dimensional CAT(0) cube complex with the piecewise ℓ8 metric, it becomes an injective metric space. The new metric we construct is weakly roughly geodesic and has the property that balls are coarsely median-convex; see Theorem 2.13. We then prove a hierarchical generalisation of a very nice result of Chepoi, Dragan, and Vaxès [CDV17] about pairwise close subsets of hyperbolic spaces. Combining this with work of Russell, Spriano, and Tran [RST18] enables us to deduce a coarse Helly property for balls. Theorem A (Proposition 2.16, Corollary 3.6). Let pX, Sq be a hierarchically hyperbolic space with metric d. There exists a metric σ on X such that pX,σq is coarsely injective and quasi-isometric to pX, dq. Moreover, σ is invariant with respect to the automorphism group of pX, Sq. Our second result relates the class of coarsely injective spaces to that of strongly shortcut spaces. A metric space has uniformly bounded geometry if, for any r ą 0, there exists a uniform Nprq P N such that every ball of radius r contains at most Nprq points. Theorem B (Theorem 4.2). Every coarsely injective metric space of uniformly bounded geometry is strongly shortcut. Huang and Osajda proved that weak Garside groups of finite type and Artin groups of FC-type are Helly [HO19], so we have the following corollary of Theorem B. Corollary C. Weak Garside groups of finite type and Artin groups of FC-type are strongly shortcut. Combining Theorem A with Theorem B, we deduce the following. Corollary D. Every hierarchically hyperbolic space admits a roughly geodesic metric in its quasi-isometry class that satisfies the strong shortcut property. In fact, in the case of hierarchically hyperbolic groups, the metric we construct is equivariant, by the “moreover” statement of Theorem A (also see Remark 3.9). We therefore have that every hierarchically hyperbolic group acts properly cocompactly on a coarsely injective space, and any group admitting such an action is a strongly shortcut group. Moreover, these three classes can be distinguished. Indeed, the p3, 3, 3q Coxeter triangle group is strongly shortcut but not coarsely injective [Hod20a]; and type-preserving ` uniform lattices in thick buildings of type Cn are coarsely injective [CCG 20], but they cannot be hierarchically hyperbolic groups, because they do not admit any nonelementary actions on hyperbolic spaces [Hae20]. 3 1.3. Metric consequences. We now describe some of the consequences of Theorem A for hierarchically hyperbolic spaces. Recall that a quasigeodesic bicombing on a metric space pX,σq is a map γ : X ˆ X ˆ r0, 1sÑ X such that, for each distinct pair a, b P X, t the map r0,σpa, bqs Ñ X given by t ÞÑ γa,b σpa,bq is a quasigeodesic from a to b with uniform constants. ´ ¯ There are various fellow-travelling conditions that a bicombing may enjoy. We say that a bicombing is roughly conical if 1 1 1 1 DC ě 0, @a, b, a , b P X, @t P r0, 1s,σpγa,bptq, γa1,b1 ptqq ď p1 ´ tqσpa, a q` tσpb, b q` C, and it is roughly reversible if it satisfies the following coarse version of symmetry:

DC ě 0, @a, b P X, @t P r0, 1s,σpγa,bptq, γb,ap1 ´ tqq ď C. From the existence of conical, reversible, isometry-invariant geodesic bicombings on injective metric spaces [Lan13], we deduce the following. Corollary E (Corollary 3.7). Let pX, Sq be a hierarchically hyperbolic space. Then pX,σq admits a roughly conical, roughly reversible, quasigeodesic bicombing that is coarsely equivariant under the automorphism group of pX, Sq. More strongly, the combing lines are rough geodesics for the metric σ. In particular, this applies to Teichmüller space with either of the standard metrics, with equivariance under the action of the mapping class group. This particular application was unknown to us until comparing results with Durham, Minsky, and Sisto [DMS20]. Corollary E gives a positive answer to Question 8.1 of [EW17], as any roughly conical bicombing is coherent and expanding, in the terminology of [EW17]. Engel and Wulﬀ proved that the existence of such a bicombing has a large number of K–theoretic consequences. This positive answer also allows one to apply work of Fukaya–Oguni (see [FO20]) to deduce the coarse Baum–Connes conjecture for hierarchically hyperbolic groups. The coarse Baum–Connes conjecture is also a consequence of ﬁnite asymptotic dimension, which is a known property of uniformly proper hierarchically hyperbolic spaces [BHS17a].

1.4. Consequences for groups. We now turn to the case of hierarchically hyperbolic groups, which, as we have seen, act properly cocompactly on coarsely injective spaces. Here we describe some of the consequences of such an action. Following Alonso and Bridson [AB95], we say that a bicombing is bounded if it satisfies the following weak two-sided fellow-traveller property: 1 1 1 1 DC ě 0, @a, b, a , b P X, @t P r0, 1s,σpγa,bptq, γa1,b1 ptqq ď C maxpσpa, a q,σpb, b qq ` C. Note that if a bicombing is roughly conical, then it is bounded. A finitely generated group is said to be semihyperbolic if it has a Cayley graph that admits an equivariant bounded quasigeodesic bicombing. Among other results, Alonso and Bridson proved that semihyperbolicity implies the existence of a quadratic isoperimetric function, that the group has soluble word and conjugacy problems, and that an algebraic flat torus theorem holds [AB95]. For more 4 discussion of the consequences of semihyperbolicity we refer the reader to [BH99]. Semi- hyperbolicity was introduced as a response to Gromov’s call for a weaker form of hyperbolicity in his original essay on hyperbolic groups, and it fits into the framework of algorithmic properties developed in [ECH`92]. For example, semihyperbolicity is implied by biautomaticity, but not by automaticity. A survey can be found in [Bri19]. For hierarchically hyperbolic groups G, the freeness of the regular action of G on pG, σq allows the bicombing of Corollary E to be pulled back to the Cayley graph of G [AB95]. Corollary F (Corollary 3.11). Every hierarchically hyperbolic group is semihyperbolic. In particular, the mapping class group of a surface of finite type is semihyperbolic. The mapping class group case is also a consequence of unpublished work of Hamenstädt [Ham09], and is related to Mosher’s automaticity theorem [Mos95]. We should emphasise that the same result for mapping class groups has been obtained by rather different methods, simultaneously and independently, by Durham, Minsky, and Sisto (see [DMS20]). This will be discussed more in Section 1.6. It is well-known that mapping class groups have finitely many conjugacy classes of finite subgroups (see [Bri00]), a property that they share with hyperbolic groups. How- ever, to the authors’ knowledge, all existing proofs of this fact rely on deep results that do not generalise to other settings, such as Kerckhoff’s celebrated solution of the Nielsen realisation problem [Ker83]. It is interesting to ask whether there is a proof that avoids such powerful machinery, and indeed a more general question about hierarchically hyperbolic groups was asked in [HP19]. The question of whether all hierarchically hyperbolic groups have finitely many conjugacy classes of finite subgroups has resisted a number of attempted resolutions. The fact that hierarchically hyperbolic groups act properly cocompactly on coarsely injective spaces makes the following a simple consequence of Lang’s result about bounded actions on injective spaces [Lan13, Proposition 1.2]. Theorem G (Corollary 3.12). Hierarchically hyperbolic groups have finitely many conjugacy classes of finite subgroups. It is interesting to note that this applies in particular to many quotients of mapping class groups [BHS17a, BHMS20]. It is also a simple consequence that residually finite hierarchically hyperbolic groups are virtually torsionfree. We now summarise the consequences for hierarchically hyperbolic groups of the results described above (also see Remark 3.9 for a comment on their generality). Corollary H. Every hierarchically hyperbolic group G has the following properties. ‚ G acts properly cocompactly on a coarsely injective space. ‚ G has finitely many conjugacy classes of finite subgroups. ‚ G is semihyperbolic. In particular: – the conjugacy problem in G is soluble, and it can be solved in doubly exponential time; – any polycyclic subgroup of G is virtually abelian; – any finitely generated abelian subgroup of G is quasi-isometrically embedded; 5 – the centraliser of any finite subset of G is finitely generated, quasi-isometrically embedded, and semihyperbolic. ‚ For any ring R, if the cohomological dimension cdRpGq is finite, then cdRpGqď asdimpGq` 1 ([EW17, Theorem C]). ‚ G is a strongly shortcut group. The result about polycyclic subgroups can also be deduced from the Tits alternative for hierarchically hyperbolic groups established in [DHS17]. The other consequences are new, however. The result about the conjugacy problem extends work of Abbott and Behrstock showing that it can be solved in exponential time for Morse elements of hierarchically hyperbolic groups [AB18], and generalises the fact that, in mapping class groups, it can always be solved in exponential time [MM00, Tao13, BD14]. In the case of cubical groups, a beautiful result of Niblo and Reeves states that every cubical group is biautomatic [NR98], and semihyperbolicity is a direct consequence of this. We emphasise, though, that the class of hierarchically hyperbolic groups is considerably larger than just cubical groups and mapping class groups.

1.5. Bounded packing. The bounded packing property for subgroups of finitely generated groups was introduced as a metric abstraction of tools used by several authors to prove intersection properties of subgroups of hyperbolic groups [GMRS98, RS99], and in turn as a step- ping stone towards ensuring cocompactness of the cube complex associated with a finite collection of quasiconvex codimension–1 subgroups [Sag97, NR03, HW14]. We recall the definition in Section 3; see [HW09, HW14] for more motivation and background. The prototypical example is that of a quasiconvex subgroup of a hyperbolic group. That such subgroups have bounded packing was first established by Gitik, Mitra, Rips, and Sageev, using compactness of the boundary [GMRS98], and another proof was given by Hruska–Wise, using induction on the height of the subgroups [HW09]. More general examples have been provided by Antolín, Mj, Sisto, and Taylor, who use induction on height to show that finite collections of stable subgroups in any finitely generated group have bounded packing [AMST19]. Stable subgroups were introduced by Durham and Taylor [DT15], and they are always hyperbolic. More generally, Morse subgroups were introduced independently by Tran [Tra19] and Genevois [Gen20], and the notion is implicit in earlier work of Sisto [Sis16]. Notably, Tran proved that any finite collection of Morse subgroups has bounded packing [Tra19, Theorem 1.2], again by using induction on height. Theorem I (Corollary 3.13). Every finite collection of hierarchically quasiconvex subgroups of a group that is a hierarchically hyperbolic space (in particular, of any hierarchically hyperbolic group) has bounded packing. For many groups that are HHSs (including all HHGs), every stable subgroup is hierarchically quasiconvex [ABD17, RST18]. Theorem B also applies to subsurface stabilisers in the mapping class group, which are neither Morse nor stable. See Section 3.1 for the definition of hierarchical quasiconvexity. Our proof of this result is purely geometric. It relies on a very strong result for quasiconvex subsets of hyperbolic spaces that was proved by Chepoi, Dragan, and Vaxès 6 [CDV17]; we state it as Theorem 3.4. Their theorem does not seem to have garnered the notice it deserves in geometric group theory. For instance, it yields what appears to be the simplest and most natural proof of bounded packing for quasiconvex subgroups of hyperbolic groups. One case of our hierarchical generalisation of their result can be stated as follows. Theorem J (Theorem 3.5). Let X be a hierarchically hyperbolic space, and let Q be a finite collection of hierarchically quasiconvex subsets of X. If every pair of elements of Q is r–close, then there is a point of X that is R–close to every element of Q, where R does not depend on the cardinality of Q. 1.6. Comparison to the work of Durham, Minsky, and Sisto [DMS20]. Let us now say a few words about the difference between the present article and the work of Durham, Minsky, and Sisto [DMS20]. As noted, both articles independently prove that mapping class groups are semihyperbolic, but the approaches differ greatly. In both cases, this fact is deduced from a stronger statement in a more general setting, but those two statements are very different in flavour. Their results hold for hierarchically hyperbolic spaces with the extra assumption of colorability, and they deduce interesting corollaries about bicombings on the Teichmüller space with the Teichmüller metric, and the existence of barycentres. These results are also consequences of Theorem A and Corollary E. Our construction is built on the fact that intervals in hierarchically hyperbolic spaces can be approximated by finite CAT(0) cube complexes (proved in [BHS17c]). The main result of Durham, Minsky, and Sisto is that these approximations are furthermore stable, meaning that a small change in the endpoints of the interval induces a small change in the approximating CAT(0) cube complex. This stability result may prove extremely useful for other purposes. If we want to compare the bicombing we obtain to the one from [DMS20] in the simplest case of a CAT(0) cube complex, our bicombing looks like the geodesic CAT(0) bicombing, whereas their bicombing is more similar to (but not the same as) Niblo–Reeves normal cube paths [NR98]. One notable difference is that our bicombing is roughly conical and their bicombing is merely bounded, which is not enough to deduce the consequences of Section 1.3. On the other hand, their bicombing paths are known to be hierarchy paths, whilst ours are not. 1.7. Structure of the article. In Section 2, we recall basic definitions of coarse median spaces, and we explain the extra property we need, a stronger approximation of median intervals by CAT(0) cube complexes. We then define a new distance, and we prove that it is quasi-isometric to the original one, is weakly roughly geodesic, and its balls are coarsely median-convex. In Section 3, we treat hierarchical hyperbolicity, and prove that hierarchically quasiconvex subsets satisfy a coarse version of the Helly property. We use this to show that the new distance makes hierarchically hyperbolic spaces coarsely injective, and deduce semihyperbolicity of hierarchically hyperbolic groups. We also show that hierarchically quasiconvex subgroups have bounded packing. In Section 4, we recall the definition of a strongly shortcut group, and prove that coarse injectivity implies the strong shortcut property. 7 Acknowledgements: We thank Mark Hagen and Anthony Genevois for many helpful comments and suggestions, and Victor Chepoi, Alexander Engel and Damian Osajda for interesting discussions. We would like to thank Matthew Durham, Yair Minsky, and Alessandro Sisto for friendly discussions about our their work and ours. We thank the anonymous referees for their careful and thorough readings of the article. The first named author was supported by the french grant ANR-16-CE40-0022-01 AGIRA. The second named author was supported by the ERC grant GroIsRan.

2. Coarse median spaces with quasicubical intervals 2.1. Background on coarse median spaces. Coarse median spaces, defined by Bowditch in [Bow13], are a generalisation of CAT(0) cube complexes and Gromov-hyperbolic spaces, and the class is rich enough to encompass mapping class groups of finite type surfaces. The general idea is to associate to every triple of points in the space a point that satisfies the axioms of a usual median up to controlled error. This point will be called the coarse median. Let us recall here that a median µ : X3 Ñ X on a set X is a map satisfying (where we write equivalently µpx,y,zq or µx,y,z to increase readability): ‚ µpx,y,zq is symmetric in x,y,z, ‚@x,y P X,µpx,x,yq“ x and ‚@a, b, x, y, z P X,µpa, b, µx,y,zq“ µpµa,b,x,µa,b,y, zq. The pair pX,µq is called a median algebra. The rank of pX,µq is the supremum of all ν P N such that there exists an injective median homomorphism from the ν–cube t0, 1uν into X. Every finite median algebra can be seen as the 0-skeleton of a CAT(0) cube complex (see [Che00, Rol98]). Let pX, dq be a metric space. For any x,y P X, let Idpx,yq “ tz P X | dpx, zq`dpz,yq “ dpx,yqu denote the interval between x and y. The metric space pX, dq is called metric median if @x,y,z P X,Idpx,yqX Idpy, zqX Idpx, zq is a singleton, say tµpx,y,zqu. In this case, µ defines a median on X. Examples of median metric spaces include trees, 1–skeletons of CAT(0) cube complexes with the combinatorial distance, and L1 spaces. In a Gromov-hyperbolic space X, the three intervals joining three points may not intersect precisely in a singleton, but by definition they do coarsely intersect with uniformly bounded diameter. This suggests defining a map X3 Ñ X that satisfies the axioms of a median up to bounded error. This is made precise by the following definition due to Bowditch [Bow13], generalising the centroid defined for mapping class groups in [BM11]. Definition 2.1 (Coarse median space). Let pX, dq be a metric space. A map µ : X3 Ñ X is called a coarse median if there exists h : N Ñ p0, `8q such that ‚ For all a, b, c, a1, b1, c1 P X, we have dpµpa, b, cq,µpa1, b1, c1qq ď hp0qpdpa, a1q ` dpb, b1q` dpc, c1qq ` hp0q. ‚ For each finite non-empty set A Ă X, with |A|ď n, there exists a finite median algebra pQ,µQq and maps π : A Ñ Q, λ : Q Ñ X such that for every α, β, γ P Q, 8 we have dpλµQpα, β, γq,µpλα, λβ, λγqq ď hpnq, and for every a P A, we have dpa, λpπpaqqq ď hpnq. We say that the triple pX,µ,dq is a coarse median space. If Q can always be chosen to have rank at most ν, we say that µ has rank at most ν. As with median algebras, we shall write µa,b,c “ µpa, b, cq interchangeably. Note that we are also free to assume that µpa, b, cq is symmetric in a, b, c, and that µpa, a, bq“ a [Bow13, p.73]. We now recall the definitions of intervals and coarse convexity in coarse median spaces. Definition 2.2 (Median interval). For a pair of points a, b P X, the median interval between a and b is defined as ra, bs “ tµpa, b, xq | x P Xu. Definition 2.3 (Coarse median-convexity). For a constant M ě 0, a subset Y of X is said to be M–coarsely median-convex if dpY,µpx,y,y1qq ď M for all y,y1 P Y, x P X. We finish by introducing some terminology. Definition 2.4 (Weakly roughly geodesic). Recall that a metric space pX, dq (or, more briefly, the metric d) is called roughly geodesic if there exists a constant Cd ě 0 such that, for any a, b P X, there exists a p1,Cdq–quasi-isometric embedding of the interval f : r0, dpa, bqs Ñ X such that fp0q“ a and fpdpa, bqq “ b. We say that a metric space pX, dq is called weakly roughly geodesic if there exists a 1 constant Cd ě 0 such that, for any a, b P X and any nonnegative r ď dpa, bq, there is a 1 1 point c P X with |dpa, cq´ r|ď Cd and dpa, cq` dpc, bqď dpa, bq` Cd. Remark 2.5. Note that every roughly geodesic metric space is weakly roughly geodesic. 1 Moreover, any metric space pX, dq that is weakly roughly geodesic with constant Cd is 1 1 necessarily p4Cd, 4Cdq–quasigeodesic. Indeed, given x,y P X, one can repeatedly take r “ 1 3Cd in the definition of weak rough geodesicity to get a sequence x “ w0, w1, . . . , wn “ y 1 1 1 such that dpwi, wi`1q P r2Cd, 4Cds and dpwi,yqď dpwi´1,yq´ Cd, and the points of this sequence form a quasigeodesic from x to y. 2.2. Construction of a new metric. Let pX,µ,dq be a coarse median space. Following Bowditch’s construction of an injective metric on a median metric space [Bow20], we shall define a new metric σ on X. Definition 2.6 (Contraction). For a constant K ě 0, a map Φ : X Ñ R is called a K–contraction if: ‚ Φ is p1, Kq–coarsely Lipschitz, i.e. @a, b P X, |Φpxq´ Φpyq| ď dpx,yq` K. ‚ Φ is a K–quasi-median homomorphism, i.e.

@a, b, c P X, |Φpµpa, b, cqq ´ µRpΦpaq, Φpbq, Φpcqq| ď K,

where µR denotes the standard median on R. Deﬁnition 2.7 (New metric). For K ą 0, we deﬁne a new metric σ on X as follows. Given a, b P X, let σpa, bq denote the supremum of all r ě 0 such that there exists a K–contraction Φ : X Ñ R such that Φpaq“ 0 and Φpbq“ r. 9 The assumption that K is nonzero is needed to ensure that σ separates points in the setting of coarse median spaces. In the special case where X is a CAT(0) cube complex, we may take K “ 0. More precisely, if X is a CAT(0) cube complex endowed with the piecewise ℓp length metric for p P t1, 2, 8u for instance, then the new metric σ for K “ 0 is the piecewise ℓ8 length metric on X. Lemma 2.8. The function σ is a metric on X. Proof. Let a, b P X be distinct. Consider the map Φ : X Ñ t0, Ku that sends b to K and everything else to 0. It is a K–contraction, and so σpa, bqě K ą 0. The proof of the triangle inequality is identical to [Bow20, Lemma 3.1]. For the reader’s convenience, we repeat it here. Let a, b, c P X. For each r ă σpa, bq there exists a K–contraction Φr : X Ñ R such that |Φrpaq´ Φrpbq| ě r. We certainly have σpa, cq ě |Φrpcq´ Φrpaq| and σpb, cq ě |Φrpbq´ Φrpcq|, so

ě supt|Φrpbq´ Φrpaq| | r ă σpa, bqu “ σpa, bq. Remark 2.9. Although the construction of σ depends on the choice of a positive constant K, the actual choice of K will not matter to us here. If K1 ă K2, then any K1– contraction is automatically a K2–contraction, so σK1 ď σK2 . On the other hand, if Φ K1 K1 2 1 is a K –contraction, then K2 Φ is a K –contraction, so σK1 ě K2 σK2 . Thus any two choices of K give biLipschitz metrics. We record the following simple consequence of the definition of σ. Lemma 2.10. If a group G is acting isometrically on a coarse median space pX,µ,dq by median isometries, in the sense that gµpx,y,zq“ µpgx,gy,gzq for all g P G, x,y,z P X, then the induced action of G on pX,µ,σq is isometric. Proof. For any g P G and x,y P X, if Φ is a K–contraction with Φpxq“ 0 and Φpyq“ r, then Φ1 “ Φg´1 is a K–contraction with Φ1pgxq“ 0 and Φ1pgyq“ r. In order to help understand the metric σ, we shall work with coarse median spaces that have the following property, which is a strengthening of the second axiom of coarse median spaces for sets A “ ta, bu with cardinality 2. We require an approximation of the entire median interval ra, bs with uniform constants, and also that the comparison map is a quasi-isometry and not just coarsely invertible. Definition 2.11 (Quasicubical intervals). Let pX,µ,dq be a coarse median space. We say that it has quasicubical intervals if it has finite rank ν and there exists κ ě 1 such that the following hold. For every a, b P X, there exists a finite CAT(0) cube complex Q 1 of dimension at most ν, endowed with the ℓ metric dQ and the median µQ, such that there exists a map λ : Q Ñ ra, bs satisfying: ‚ λ is a pκ, κq–quasi-isometry, i.e. λ is κ–coarsely onto and 1 @α, β P Q, d pα, βq´ κ ď dpλpαq, λpβqq ď κd pα, βq` κ; κ Q Q ‚ λ is a κ–quasi-median homomorphism, i.e.

@α, β, γ P Q, dpλpµQpα, β, γqq,µpλpαq, λpβq, λpγqq ď κ. 10 Obviously this is satisfied by finite dimensional CAT(0) cube complexes, or indeed by any space with a global quasi-median quasi-isometry to a CAT(0) cube complex. Proposition 2.12. Hierarchically hyperbolic spaces have quasicubical intervals, as do coarse median spaces satisfying the axioms (B1)-(B10) in [Bow19a]. Proof. In hierarchically hyperbolic spaces, the notion of median intervals used here coincides coarsely with the hierarchically quasiconvex hull of a pair of points defined in [BHS19], by [RST18, Corollary 5.12] and [Bow19b, Lemma 8.1]. The first statement is thus a special case of [BHS17c, Theorem 2.1]. The second statement is exactly [Bow19a, Theorem 1.3]. As noted by Bowditch, every hierarchically hyperbolic space satisfies the axioms (B1)- (B10) in [Bow19a]. It is not known whether all cocompact cube complexes can be given a structure that satisfies these axioms. We can now state the main result of this section. It sums up Lemma 2.10, Proposi- tion 2.16, Proposition 2.21, and Lemma 2.23, and the proof is split over the next three subsections. Theorem 2.13. Assume that the coarse median space pX,µ,dq has quasicubical intervals and is roughly geodesic. The metrics σ and d are quasi-isometric, σ is weakly roughly geodesic, and balls for σ are uniformly coarsely median-convex. Moreover, σ is invariant under the group of median isometries of pX,µ,dq. 2.3. The metrics d and σ are quasi-isometric. Here we shall prove that the new distance σ is quasi-isometric to the original distance d. We need the following technical result for coarse median spaces, which is a special case of Lemmas 2.18 and 2.19 of [NWZ19].

Lemma 2.14. In any coarse median space pX,d,µq, there exists a constant H5 ě 0 such that the following inequalities hold for any a, b, x, y, z P X.

dpµpa, b, µx,y,zq,µpµa,b,x,µa,b,y, zqq ď H5

dpµpa, b, µx,y,zq,µpµa,b,x,µa,b,y,µa,b,zqq ď H5. We will now prove that, up to multiplicative and additive constants, one can restrict to contractions defined on the interval between two points for the definition of σ. Lemma 2.15. For each a, b P X, let σ1pa, bq denote the supremum of all r ě 0 such that there exists a K–contraction Φ1 : ra, bs Ñ R for which Φ1paq “ 0 and Φ1pbq “ r. There exists L ě 1 such that, for each a, b P X, we have σpa, bqď σ1pa, bqď Lσpa, bq. Proof. It is immediate that σpa, bqď σ1pa, bq. Consider r ě 0 and a K–contraction Φ1 : ra, bs Ñ R such that Φ1paq “ 0 and Φ1pbq “ r. Define Φ : X Ñ R by c ÞÑ Φ1pµpa, b, cqq. Since the map c ÞÑ µpa, b, cq is php0q, hp0qq–coarsely Lipschitz and Φ1 is p1, Kq–coarsely Lipschitz, we deduce that Φ is php0q, hp0q` Kq–coarsely Lipschitz. Now let x,y,z P X. According to Lemma 2.14 we have

dpµpa, b, µx,y,zq,µpµa,b,x,µa,b,y,µa,b,zqq ď H5. 11 Hence, since Φ1 is p1, Kq–coarsely Lipschitz, 1 1 |Φ pµpa, b, µx,y,zqq ´ Φ pµpµa,b,x,µa,b,y,µa,b,zqq| ď H5 ` K. But Φ1 is also a K–quasi-median homomorphism, and so 1 1 1 1 |Φ pµpµa,b,x,µa,b,y,µa,b,zqq ´ µRpΦ pµa,b,xq, Φ pµa,b,yq, Φ pµa,b,zqq| ď K.

Combining these and recalling the deﬁnition of Φ enables us to conclude that |Φpµx,y,zq´ hp0q H5 µRpΦpxq, Φpyq, Φpzqq| ď H5 ` 2K. Thus, if we set L “ maxthp0q, 1 ` K , 2 ` K u, then 1 1 we have that L Φ is a K–contraction, and so σ pa, bqď Lσpa, bq. We can now deduce that σ is quasi-isometric to d in the setting of Theorem 2.13. Proposition 2.16. If pX,µ,dq has quasicubical intervals, then d and σ are quasi-isometric. Proof. Fix a, b P X. First of all, since any K–contraction is p1, Kq–coarsely Lipschitz, we have σpa, bqď dpa, bq` K. According to the quasicubicality of intervals, there exists a ﬁnite CAT(0) cube complex Q of dimension at most ν, and a map λ : pQ, dQq Ñ ra, bs that is a pκ, κq–quasi-isometry and a κ–quasi-median homomorphism. Then λ has a quasi-inverse π : ra, bs Ñ pQ, dQq that is a pκ1, κ1q–quasi-isometry and a κ1–quasi-median homomorphism, where κ1 is a constant depending only on κ and hp0q. Note that we shall in fact use Q to denote the vertex set, dQ to denote the combinatorial 1 (piecewise ℓ ) distance on Q, and µQ to denote the median on Q. Let us denote by σQ 8 the piecewise ℓ distance on Q: we have σQ ď dQ ď νσQ.

Since Q is a CAT(0) cube complex, there exists a 0–contraction ΦQ : pQ, dQqÑ Z such that ΦQpπpaqq “ 0 and ΦQpπpbqq “ σQpπpaq,πpbqq (see [Bow20, §7], [BvdV91, Cor. 2.5]). 1 mint1,Ku R 1 1 Let us consider Φ “ κ1 ΦQπ : ra, bs Ñ . Since π is a pκ , κ q–quasi-isometry and 1 ΦQ is 1–Lipschitz, we deduce that Φ is p1, Kq–coarsely Lipschitz. Furthermore, for every x,y,z P ra, bs, we have:

|ΦQπpµx,y,zq´ µRpΦQπpxq, ΦQπpyq, ΦQπpzqq|

ď |ΦQπpµx,y,zq´ ΦQpµQpπpxq,πpyq,πpzqqq|

` |ΦQpµQpπpxq,πpyq,πpzqqq ´ µRpΦQπpxq, ΦQπpyq, ΦQπpzqq| 1 ď dQpπpµx,y,zq,µQpπpxq,πpyq,πpzqqq ď κ , so Φ1 is K–quasi-median. The map Φ1 is therefore a K–contraction on ra, bs, and Φ1paq “ 0 and Φ1pbq “ mint1,Ku mint1,Ku κ1 σQpπpaq,πpbqq ě νκ1 dQpπpaq,πpbqq. Using Lemma 2.15, we deduce that νκ1L 1 1 dQpπpaq,πpbqq ď mint1,Ku σpa, bq. But π is a pκ , κ q–quasi-isometry, so we also have 1 1 dQpπpaq,πpbqq ě κ1 dpa, bq´ κ . In conclusion, we have mint1, Ku mint1, Ku dpa, bq´ ď σpa, bq ď dpa, bq` K νκ12L νL for all a, b P X. 12 2.4. The metric σ is weakly roughly geodesic. Recall that pX,µ,dq is a coarse median space with corresponding function h, that X has quasicubical intervals (though this will only be used for Proposition 2.21 in this section), and that the metric d is Cd–roughly geodesic. We shall prove that the new metric σ is weakly roughly geodesic (see Deﬁnition 2.4). This will be the most diﬃcult part of the proof of Theorem 2.13.

Let a, b P X, let E be a small positive constant, and consider K–contractions Φ1 : X Ñ r0, rs and Φ2 : X Ñ rr, r ` ss (for some r, s ě E) such that Φ1paq ď E and Φ2pbqě r ` s ´ E. We want to ﬁnd a criterion to ensure that we can combine Φ1 and Φ2 into a contraction Φ such that Φpaq “ 0 and pr ` sq´ Φpbq is bounded above by some constant.

Lemma 2.17. Assume that a, b, Φ1, Φ2, r, s, E are as above. Let D “ hp0qp3K ` 4Cdq` 3K ` hp0q. If t P r0, mintr, su´ D ` K ´ Es is such that the sets

Z1 “ tx P ra, bs | Φ1pxqď r ´ tu and Z2 “ tx P ra, bs | Φ2pxqě r ` tu are disjoint, then σpa, bqě r ` s ´ 2t ´ 2D ´ 2E.

m Proof. For m P t0, 1, 2u, let us write Y1 “ tx P ra, bs | Φ1pxq ď r ´ t ´ D ` mKu and m m1 m2 Y2 “ tx P ra, bs | Φ2pxqě r ` t ` D ´ mKu. Note that if m1 ă m2, then Yi Ă Yi . 2 2 Claim 1: dpY1 ,Y2 qě D ´ 3K 2 2 2 Proof of Claim 1: Let x1 P Y1 and x2 P Y2 . Since Y2 Ă Z2, we have x2 R Z1, so Φ1px2qą r ´ t. We also have Φ1px1qď r ´ t ´ D ` 2K, so |Φ1px1q´ Φ1px2q| ě D ´ 2K. As Φ1 is p1, Kq–coarsely Lipschitz, we have |Φ1px1q´Φ1px2q| ď dpx1,x2q`K, and hence dpx1,x2qě D ´ 3K. ♦

1 1 Claim 2: dpY1 ,Y2 qě 3K ` 4Cd. 1 1 Proof of Claim 2: Let x1 P Y1 and x2 P Y2 , and set y1 “ µpa, b, x1q P ra, bs and y2 “ µpa, b, x2q P ra, bs. We know that Φ1py1qď µRpΦ1paq, Φ1pbq, Φ1px1qq ` K. We also have Φ1paqď E by assumption, and Φ1px1qď r ´ t ´ D ` K. As this latter quantity is at least E, µRpΦ1paq, Φ1pbq, Φ1px1qq ď r ´ t ´ D ` K. Hence Φ1py1qď r ´ t ´ D ` 2K, 2 2 so y1 P Y1 . A similar argument shows that y2 P Y2 . According to Claim 1, dpy1,y2q ě D ´ 3K. Since µ is php0q, hp0qq–coarsely Lipschitz with respect to each variable, we have dpy1,y2q ď hp0qdpx1,x2q` hp0q, so dpx1,x2q ě 0 3 0 dpy1,y2q´hp q D´ K´hp q ♦ hp0q ě hp0q “ 3K ` 4Cd, as desired.

0 0 Claim 3: The set tµx,y,z | x,y P Y1 , z P Xu is disjoint from Y2 , and tµx,y,z | x,y P 0 0 Y2 , z P Xu is disjoint from Y1 . 0 Proof of Claim 3: Fix x,y P Y1 and z P X. Since Φ1pxq, Φ1pyqď r´t´D, we deduce that µRpΦ1pxq, Φ1pyq, Φ1pzqq ď r ´ t ´ D, and it follows that Φ1pµx,y,zqď r ´ t ´ D ` K, 1 1 1 so µx,y,z P Y1 . Because we showed in Claim 2 that dpY1 ,Y2 qě 3K ` 4Cd ą 0, we know 1 0 that µx,y,z R Y2 , and, in particular, µx,y,z R Y2 . The other case is similar. ♦ 13 0 0 Write Y “ X r pY1 Y Y2 q, and consider Φ : X Ñ r0, r ` s ´ 2t ´ 2Ds deﬁned by: 0 If x P Y1 then Φpxq“ Φ1pxq. 0 If x P Y2 then Φpxq“ Φ2pxq´ 2t ´ 2D. If x P Y then Φpxq“ r ´ t ´ D. We have Φpaqď E and Φpbqě r`s´2t´2D´E, so if we prove that Φ is a K–contraction, then we may deduce that σpa, bqě r ` s ´ 2t ´ 2D ´ 2E, the desired conclusion. Claim 4: Φ is p1, Kq–coarsely Lipschitz. 0 Proof of Claim 4: Notice that Φ coincides on Y1 Y Y with the composition of Φ1 : X Ñ r0, rs with the 1–Lipschitz map mt “ minp¨, r ´ t ´ Dq : r0, rs Ñ r0, r ´ t ´ Ds. 0 Hence, if x,y P Y1 Y Y , then |Φpxq´ Φpyq| ď |Φ1pxq´ Φ1pyq| ď dpx,yq` K. A similar 0 argument involving a maximum function applies if x,y P Y2 Y Y . 0 0 Now suppose that x P Y1 and y P Y2 . Since d is Cd–roughly geodesic, there is a p1,Cdq–quasi-isometric embedding f : r0, dpx,yqs Ñ X with fp0q “ x, fpdpx,yqq “ y. 0 0 For any ε ą 0, there exists τ such that fpτq P Y1 but fpτ ` δq R Y1 for any δ ą ε. 0 (Were f continuous, we could take ε “ 0 and use the maximal τ with fpτq P Y1 .) Write z1 “ fpτq. We have dpx, z1q` dpz1,yqď dpx,yq` Cd and Φ1pz1qď r ´ t ´ D. Moreover, for any δ ą ε we have

Φ1pz1q ě Φ1pfpτ ` δqq ´ pdpfpτq,fpτ ` δqq ` Kq ą r ´ t ´ D ´ pδ ` Cd ` Kq, 0 and so Φ1pz1qě r ´ t ´ D ´ Cd ´ K ´ ε. We can now similarly construct z2 P Y2 such that dpz1, z2q` dpz2,yqď dpz1,yq` Cd and r ` t ` D ď Φ2pz2qď r ` t ` D ` Cd ` K ` ε. With these, we can compute

|Φpxq´ Φpyq| ď |Φpxq´ Φpz1q| ` |Φpz1q´ Φpz2q| ` |Φpz2q´ Φpyq| “ |Φ1pxq´ Φ1pz1q| ` |Φ1pz1q ´ pΦ2pz2q´ 2t ´ 2Dq| ` |Φ2pz2q´ Φ2pyq|

ď pdpx, z1q` Kq ` p2Cd ` 2K ` 2εq ` pdpz2,yq` Kq

ď pdpx,yq` Cd ´ dpz1,yq` Kq ` p2Cd ` 2K ` 2εq ` pdpz1,yq` Cd ´ dpz1, z2q` Kq

“ dpx,yq´ dpz1, z2q` 4K ` 4Cd ` 2ε ď dpx,yq` K ` 2ε, 0 0 1 1 where the last line comes from Claim 2: dpz1, z2qě dpY1 ,Y2 qě dpY1 ,Y2 qě 3K ` 4Cd. This is suﬃcient, because ε can be taken to be arbitrarily close to 0. ♦ Claim 5: Φ is K–quasi-median. 0 Proof of Claim 5: As noted in the proof of Claim 4, on Y1 Y Y we have Φ “ mtΦ1. 0 As mt is a median homomorphism with respect to µR, if x,y,z P Y1 Y Y , then

|Φpµx,y,zq´ µRpΦpxq, Φpyq, Φpzqq| “ |mtΦ1pµx,y,zq´ µRpmtΦ1pxq,mtΦ1pyq,mtΦ1pzqq|

ď |Φ1pµx,y,zq´ µRpΦ1pxq, Φ1pyq, Φ1pzqq| ď K, 0 and similarly if x,y,z P Y2 Y Y . 0 0 Assume now that x,y P Y1 and z P Y2 . We ave that both Φpxq and Φpyq are at most 0 r ´t´D. Moreover, Φpzq“ Φ2pzq´2t´2D ě r ´t´D, and the fact that z R Y1 implies that Φ1pzqą r ´ t ´ D. Thus µRpΦpxq, Φpyq, Φpzqq “ µRpΦ1pxq, Φ1pyq, Φ1pzqq ď r ´ t ´ D. As Φ1 is K–quasi-median, we deduce that |µRpΦpxq, Φpyq, Φpzqq ´ Φ1pµx,y,zq| ď K. 14 0 According to Claim 3, we know that µx,y,z R Y2 , and so Φpµx,y,zq “ mtΦ1pµx,y,zq. But µRpΦpxq, Φpyq, Φpzqq ď r´t´D, so we conclude that |µRpΦpxq, Φpyq, Φpzqq´Φpµx,y,z q| ď 0 0 K. A similar argument applies when x,y P Y2 and z P Y1 . 0 0 Assume ﬁnally that x P Y1 , y P Y , and z P Y2 . Since Φpxq “ Φ1pxq ď r ´ t ´ D, Φpyq“ r´t´D and Φpzq“ Φ2pzq´2t´2D ě r´t´D, we have µRpΦpxq, Φpyq, Φpzqq “ 0 r ´ t ´ D. If µx,y,z P Y , then Φpµx,y,zq“ r ´ t ´ D “ µRpΦpxq, Φpyq, Φpzqq. If µx,y,z P Y1 , then Φpµx,y,zq“ Φ1pµx,y,zqě µRpΦ1pxq, Φ1pyq, Φ1pzqq ´ K ě r ´ t ´ D ´ K, from which it follows that |µRpΦpxq, Φpyq, Φpzqq ´ Φpµx,y,zq| ď K. A similar argument applies if 0 µx,y,z P Y2 . We have shown that |µRpΦpxq, Φpyq, Φpzqq ´ Φpµx,y,zq| ď K in all cases. ♦ We have proved that Φ is a K–contraction. As stated above, this shows that σpa, bqě |Φpaq´ Φpbq| ě r ` s ´ 2t ´ 2D ´ 2E. Recall that the convex hull HullpAq of a subset A of a CAT(0) cube complex Q is the smallest convex subcomplex of Q containing A. Equivalently, it is the smallest subcomplex that contains A and is median convex, in the sense that µpq, a, bq P HullpAq whenever a, b P HullpAq. We regard a subset of Qp0q as convex if the full subcomplex spanned by it is convex. We need the following iterative description convex hulls in CAT(0) cube complexes.

Lemma 2.18. Let Q be a CAT(0) cube complex of dimension at most ν, and let µQ : 0 3 0 0 Qp q Ñ Qp q denote the median. Given A Ă Qp q, set A0 “ A, and for each i P N, let p0q p0q Ai`1 “ µQpQ , Ai, Aiq “ tµQpx, a, bq | a P Ai, b P Ai,x P Q u. 1 Then Aν1 “ HullpAq, where ν “ maxt1,ν ´ 1u. Note that the constant ν1 is probably far from optimal. However, ν1 does depend on A and Q. For example, if A is the star of a vertex in a ν–cube, then it can be seen that 1 the optimal value of ν is rlog2 νs in this case. Proof. The result is trivial if A is convex. Otherwise, fix x P HullpAq r A, and let H be the collection of hyperplanes of HullpAq that are adjacent to x. For each H P H, let p0q ` ´ ` Q “ H \ H denote the partition defined by H, where x P H . Let tH1,...,Hnu be a maximal pairwise crossing family in H. We have n ď ν. For each i, let Hi denote the set of elements of H that are disjoint from Hi, together with Hi. An important ´ ` r observation is that Hi Ă H whenever H P Hi tHiu. If n “ 1, then x is a cut-point or leaf of HullpAq, so taking any a P AXH`, b P AXH´ gives x “ µpx, a, bq, and we are done. ´ ´ So suppose that n ě 2. If for every a P A X H1 we have a P H2 , then for every ` ` ´ ` b P A X H2 we have b P H1 , so if we take z1 P A X H1 and z2 P A X H2 , then µpx, z1, z2q P H` X H1` for every H P H1, H1 P H2. We can reason similarly if every ´ ´ ´ ` element of A X H2 lies in H1 . Otherwise there exist z1 P A X H1 X H2 and z2 P ` ´ ` 1` 1 A X H1 X H2 , and we again have µpx, z1, z2q P H X H for every H P H1, H P H2. Let y1 “ µpx, z1, z2q P A1. ` We proceed inductively. Suppose that we have yi P Ai such that yi P H for all H P jďi`1 Hj. Let zi`2 be any point of A that is separated from yi by Hi`2, and set y 1 “ µpx,y , z 2q. Since x,y P H` for every H P H , the same is true of y 1, i` Ť i i` i jďi`1 j i` 15 Ť ` and since yi and zi`2 lie on opposite sides of Hi`2, we also have that yi`1 P H for all H P Hi`2. By this procedure, we obtain yn´1 P An´1 X HullpAq that is not separated from x by any hyperplane of HullpAq, so we must have yn´1 “ x. In order to apply Lemma 2.17, we focus on contractions on CAT(0) cube complexes. Recall that a chain of hyperplanes is an ordered sequence pH1,...,Hnq of pairwise disjoint hyperplanes such that Hj separates Hi from Hk whenever i ă j ă k. Lemma 2.19. Let Q be a CAT(0) cube complex of dimension at most ν, and let Φ : Qp0q Ñ R be a K1–quasi-median, pK1, K1q–coarsely Lipschitz map (for the ℓ1 metric) with bounded image. There exists an interval ru, vs of Z and a chain pHnquďnďv of hyperplanes of Q satisfying the following: ‚ for each vertex x in Q, there exists a unique n “ Ψpxq P ru ´ 1, vs such that: ‹ either u ď n ď v ´ 1 and x is between Hn and Hn`1, ‹ or n “ u ´ 1 and Hu separates x from Hu`1, ‹ or n “ v and Hv separates x from Hv´1, and ‚ for each vertex x in Q, we have |Φpxq´ 4K1νΨpxq| ď 4K1ν. ´1 1 p0q Proof. Fix n P Z, and consider Kn “ Φ pp2AnÁ, 2Ansq, where A “ 2K ν. Since Q is 1–connected, we know that ΦpQp0qq is 2K1–connected. In particular, the set of integers n P Z such that Kn ‰H is an interval ru ´ 1, vs. Furthermore, for each u ď n ď v ´ 1, we know that Kn disconnects Q. In the notation of Lemma 2.18 we have that, for all i ě 0, if x P pKnqi, then |Φpxq´ 1 ΦpKnq| ď K i. Indeed, this is clear for i “ 0, and if x P pKnqi`1, so that there exist 1 a, b P pKnqi with x “ µQpx, a, bq, then the fact that Φ is K –quasimedian implies that |Φpxq´ µRpΦpxq, Φpaq, Φpbqq| ď K1, yielding the claimed inequality by induction. In 1 particular, Lemma 2.18 tells us that every x P HullpKnq satisfies |Φpxq´ ΦpKnq| ď K ν. As a consequence, if n ‰ m, then the convex subcomplexes HullpKnq and HullpKmq are disjoint. Thus, for each u ď n ď v, there exists a hyperplane Hn of Q that separates HullpKn´1q from HullpKnq [Che94, Corollary 1]. For each vertex x P Qp0q, let u ´ 1 ď n ď v be such that Φpxq P p2An ´ 2A, 2Ans. Then Ψpxq is either equal to n ´ 1 or to n. So |Φpxq´ 2AΨpxq| ď 2A “ 4K1ν. Before stating the next lemma, we remark that, given any chain H of hyperplanes in a finite CAT(0) cube complex Q, there is an associated map Qp0q Ñ Z: the cube complex dual to H is a finite interval of Z, and each vertex of Q determines a consistent orientation of the hyperplanes in H. This is a special case of the restriction quotient described in [CS11], and it is clearly a median map. Conversely, any 0–contraction on Q can be realised as restriction quotient in this manner. Moreover, after a translation of Z, we may assume that the codomain is contained in N if it is bounded. Lemma 2.20. Let Q be a finite CAT(0) cube complex of dimension at most ν. Let C be a (necessarily finite up to translations of Z) family of 0–contractions on Q, i.e. each p0q Ψ P C is a map Q Ñ N given by a chain pHΨ,1,...,HΨ,nΨ q of hyperplanes of Q. Let 0 σC denote the pseudometric on Qp q defined by p0q @α, β P Q ,σC pα, βq“ max |Ψpαq´ Ψpβq|. ΨPC 16 0 Then for each α, β P Qp q and for each integer 0 ď r ď σCpα, βq, there is a vertex γ P rα, βs and contractions Ψ1, Ψ2 P C such that the following hold. (1) σCpα, γq“ r, (2) σCpα, γq “ |Ψ1pαq´ Ψ1pγq| and σCpγ, βq “ |Ψ2pγq´ Ψ2pβq|,

(3) if pH1,1,...,H1,n1 q is the maximal subchain of hyperplanes deﬁning Ψ1 that sep-

arate α from γ, and pH2,1,...,H2,n2 q is the maximal subchain of hyperplanes

deﬁning Ψ2 that separate γ from β, then pH1,1,...,H1,n1 ,H2,1,...,H2,n2 q is a chain. 0 Proof. Fix α, β P Qp q and an integer 0 ă r ă σCpα, βq. Since σC is 1–Lipschitz with respect to the combinatorial distance on Qp0q, we know that there exists γ P rα, βs such that σCpα, γq “ r. Among all possible choices, choose such γ as far away from α as possible, in the sense that: 1 1 1 1 if γ P rα, βs has σCpα, γ q“ r and γ P rα, γ s, then γ “ γ.

Let Ψ2 P C such that σCpγ, βq “ |Ψ2pγq´Ψ2pβq|. Let pH2,1,...,H2,n2 q be the maximal subchain of hyperplanes defining Ψ2 that separate γ from β, numbered from γ to β. Let H be a hyperplane of Q adjacent to γ and either equal to H2,1 or separating γ 1 from H2,1, and let γ P rα, βs be the vertex adjacent to γ such that H crosses the edge 1 rγ, γ s. First note that, since H2,1 separates γ and β, we deduce that H separates γ and β. Thus H does not separate α and γ, because γ P rα, βs. In particular, γ P rα, γ1s. Since γ is chosen as far from α as possible among points at σC–distance equal to r, and every hyperplane separating α and γ separates α and γ1, we deduce that σCpα, γ1q ą σCpα, γq“ r, so σCpα, γ1q“ σCpα, γq` 1. 1 1 Let Ψ1 P C such that σCpα, γ q “ |Ψ1pαq´ Ψ1pγ q|. Let pH1,1,...,H1,n1`1q be the maximal subchain of hyperplanes defining Ψ1 that separate α from γ1, numbered from 1 1 α to γ . Since σCpα, γ q“ σCpα, γq` 1, we know that H “ H1,n1`1 and that σCpα, γq“ |Ψ1pαq´ Ψ1pγq|. In particular, H is disjoint from H1,1,...,H1,n1 . We deduce that H separates H1,1,...,H1,n1 from H2,1,...,H2,n2 , and the conclusion follows. We can now use these lemmas to prove that, in the setting of Theorem 2.13, the metric σ is weakly roughly geodesic (Definition 2.4). Proposition 2.21. If pX,µ,dq has quasicubical intervals and is roughly geodesic, then σ is weakly roughly geodesic. Proof. Let a, b P X. Since X has quasicubical intervals, there exists a finite CAT(0) cube complex Q (with the ℓ1 metric) of dimension at most ν, and a map λ : Q Ñ ra, bs that is a pκ, κq–quasi-isometry and a κ–quasi-median homomorphism. We can therefore fix α, β P Q such that dpλpαq, aqď κ and dpλpβq, bqď κ. According to Proposition 2.16, there is a constant q ě 1 such that d and σ are pq,qq–quasi-isometric. It follows that σpλpαq, aqď qpκ ` 1q and σpλpβq, bqď qpκ ` 1q. For each K–contraction Φ : X Ñ R, the composition Φλ : Q Ñ R is a K1–quasi- median, pK1, K1q–coarsely Lipschitz map, where K1 “ K ` κ. According to Lemma 2.19, there exists a 0–contraction Ψ : Q Ñ Z such that |Φλpξq´ 4K1νΨpξq| ď 4K1ν for all ξ P Qp0q. Let C denote the set of all 0–contractions Ψ : Q Ñ Z such that there is some K–contraction Φ : X Ñ Z with |Φλpξq´ 4K1νΨpξq| ď 4K1ν for all ξ P Qp0q. 17 We shall prove that σ is weakly roughly geodesic with constant 1 1 Cσ “ 64K ν ` 4qpκ ` 1q` 4κ ` 4K ` 2D, where D is the constant from Lemma 2.17. 1 Let r P r0,σpa, bqs. If r ă Cσ, then clearly we can take c “ a for the desired point. 1 Similarly, if r ą σpa, bq´ Cσ, then we can take c “ b. Otherwise, Lemma 2.20 applied 1 r to α, β, the family C, and r “ t 4K1ν u provides a vertex γ P rα, βs and 0–contractions Ψ1, Ψ2 P C. Let c “ λpγq P ra, bs. Let us start by computing σpa, cq. By definition of the set C, for any K–contraction Φ : X Ñ R there is some Ψ P C (and vice versa: for any Ψ P C there exists a K– contraction Φ) such that |Φλpξq´ Φλpζq| ´ 4K1ν|Ψpξq´ Ψpζq| 1 1 1 ˇ ď |Φλpξq´ 4K νΨpξq| `ˇ |Φλpζq´ 4K νΨpζq| ď 8K ν ˇ ˇ holds for all ξ,ζ P Qp0q. Itˇ follows that ˇ ˇ ˇ 1 1 (1) |σpλpξq, λpζqq ´ 4K νσCpξ,ζq| ď 8K ν.

By the choice of γ, we have σCpα, γq“ r1. Thus, from (1) we obtain |σpa, cq´ r| ď |σpλpαq, λpγqq ´ 4K1νr1|` qpκ ` 1q` 4K1ν 1 1 ď 12K ν ` qpκ ` 1q ď Cσ. The aim for the rest of the proof is to confirm the second restriction on c, that σpa, cq` 1 σpc, bqď σpa, bq` Cσ. The strategy is to apply Lemma 2.17. Recall that Ψ1, Ψ2 P C are the 0–contractions provided by Lemma 2.20: they satisfy σCpα, γq “ |Ψ1pαq´ Ψ1pγq| “ r1 and σCpγ, βq “ |Ψ2pγq´ Ψ2pβq| “ s1. After translations of Z, we may also assume that Ψ1pαq “ 0, Ψ1pγq “ Ψ2pγq “ r1, and Ψ2pβq “ r1 ` s1. By definition of C, there exist K–contractions Φ1 and Φ2 on X such that |Φ1λpξq´ 0 4K1νΨ1pξq| ď 4K1ν and |Φ2λpξq´ 4K1νΨ2pξq| ď 4K1ν for all ξ P Qp q. In particular, Φ1paqď Φ1λpαq` κ ` K ď 4K1ν ` κ ` K. Moreover, by using (1) we see that 1 1 1 1 Φ2pbqě 4K νpr ` s q´ 4K ν ´ κ ´ K 1 1 “ 4K νpσCpα, γq` σCpγ, βqq ´ 4K ν ´ κ ´ K ě σpλpαq, λpγqq ` σpλpγq, λpβqq ´ 16K1ν ´ 4K1ν ´ κ ´ K ě σpa, cq` σpc, bq´ 20K1ν ´ 2qpκ ` 1q´ κ ´ K. We are now in the setting of Lemma 2.17, with E “ 20K1ν ` 2qpκ ` 1q` κ ` K and with the image of Φ2 being bounded above by σpa, cq` σpc, bq. Let us show that the assumptions of the lemma are met if we take t “ 12K1ν ` κ ` K. 1 We must first note that r ´ D ` K ´ E ě Cσ ´ D ` K ´ E ě t, and secondly that 1 σpa, cq` σpc, bq´ r ´ D ` K ´ E ě σpa, bq´ r ´ D ` K ´ E ě Cσ ´ D ` K ´ E ě t. It remains to prove that the subspaces Z1 “ tx P ra, bs | Φ1pxq ď r ´ tu and Z2 “ 0 tx P ra, bs | Φ2pxq ě r ` tu are disjoint. Fix x P ra, bs, and pick any ξ P Qp q such that dpλpξq,xqď κ. 18 If x P Z1, so that Φ1pxq ď r ´ t, then Φ1pλpξqq ď r ´ t ` κ ` K, and hence Ψ1pξq ď r´t`κ`K`4K1ν r´8K1ν 1 1 4K1ν ď 4K1ν ď r ´ 1. Similarly, if x P Z2, then Ψ2pξqě r ` 1. According to Property (3) of Lemma 2.20, the halfspace of HΨ1,r1 containing α is p0q disjoint from the halfspace of HΨ2,1 containing β. Thus, if ξ P Q , then we cannot simultaneously have both Ψ1pξq ď r1 ´ 1 and Ψ2pξq ě r1 ` 1. As a consequence, we cannot have both x P Z1 and x P Z2. This implies that Z1 X Z2 “H. The conditions of Lemma 2.17 are therefore met, and by applying it we deduce that 1 σpa, bq ě σpa, cq` σpc, bq´ 2t ´ 2D ´ 2E “ σpa, cq` σpc, bq´ Cσ. This completes the 1 proof that σ is weakly roughly geodesic with constant Cσ. 2.5. Coarse convexity of balls. To complete the proof of Theorem 2.13, it remains to show that balls in pX,σq are uniformly coarsely median-convex. Lemma 2.22. There is a constant ǫ ě 0 such that for any x,y,z P X with x P ry, zs, we have dpx,µpx,y,zqq ď ǫ. 1 Proof. According to [Bow19b, Lemma 8.1], there are constants r0 and r0 such that x 1 1 1 1 lies at distance at most r0 from a point x with dpx ,µpx ,y,zqq ď r0. Since the coarse median µ is coarsely Lipschitz, we have dpµpx,y,zq,µpx1,y,zqq ď hp0qdpx,x1q` hp0q ď 1 1 1 hp0qpr0 ` 1q. We deduce that dpx,µpx,y,zqq ď ǫ, where ǫ “ r0 ` r0 ` hp0qpr0 ` 1q. Lemma 2.23. Suppose that pX,µ,dq has quasicubical intervals and is roughly geodesic. There is a constant M such that each ball in pX,σq is M–coarsely median-convex.

Proof. Fix w P X and R ě 0. Let y, z P Bσpw, Rq. Given any a P X, we want to bound the distance from x “ µpa, y, zq to Bσpw, Rq. Let r ă σpw,xq, and let Φ : X Ñ r0, rs be a K–contraction such that Φpwq “ 0 and Φpxqě r. Lemma 2.22 tells us that dpµx,y,z,xqď ǫ, so |Φpµx,y,zq´ Φpxq| ď ǫ ` K. Since Φ is a K–quasi-median homomorphism, we have

µRpΦpxq, Φpyq, Φpzqq ě Φpµx,y,zq´ K ě Φpxq´ ǫ ´ 2K ě r ´ ǫ ´ 2K. This means that one of Φpyq and Φpzq must be at least r ´ ǫ ´ 2K, and so σpw,xq ď maxtσpw,yq,σpw, zqu ` ǫ ` 2K. This proves that x P Bσpw, R ` ǫ ` 2Kq. According to Proposition 2.21, σ is weakly 1 roughly geodesic with constant Cσ. Applying Deﬁnition 2.4 with a “ w, b “ x, and 1 1 1 1 r “ mintR´Cσ,σpw,xqu, yields a point x P Bσpw, Rq with dpx ,xqď ǫ`2K`3Cσ “ M, which shows that balls in pX,σq are M–coarsely median-convex.

3. Quasiconvexity and a coarse Helly property in HHSs The goal of this section is to prove that hierarchically quasiconvex subsets of hierarchically hyperbolic spaces satisfy a coarse version of the Helly property. Since coarsely median-convex subsets of a hierarchically hyperbolic space are hierarchically quasiconvex [RST18, Proposition 5.11], this applies in particular to balls for the metric σ constructed in Section 2, by Theorem 2.13, allowing us to deduce Theorem A. We also deduce the bounded packing property for hierarchically quasiconvex subgroups of groups that are HHSs. 19 3.1. Background on hierarchical hyperbolicity. Here we give a description of hierarchically hyperbolic spaces (HHSs) and hierarchically hyperbolic groups (HHGs). For full definitions, see [BHS19, Def. 1.1, 1.21]. Briefly, an HHS consists of a quasigeodesic space pX, dq, a constant E, and a set S, elements of which are called domains. Each domain U has an associated E–hyperbolic space CU, and the various axioms give structure for extracting information about X from these hyperbolic spaces. This includes: ‚ Each domain U has an associated E–coarsely onto, pE, Eq–coarsely Lipschitz projection map πU : X Ñ CU. ‚ S has a partial order <, called nesting, and a symmetric relation K, called or- thogonality. If U < V and V KW , then UKW . The relations Ĺ, K, and “ are mutually exclusive, and their complement, denoted &, is called transversality. ‚ There is a bound on the size of Ĺ–chains and pairwise orthogonal sets. U ‚ If U Ĺ V or U&V then there is a set ρV Ă CV of diameter at most E. V ‚ If U Ĺ V then there is also a map ρU : CV Ñ CU. If γ Ă CV is a geodesic and U V dCV pγ, ρV qą E, then diam ρU pγqď E. This last point is referred to as bounded geodesic image. For x,y P X, it is standard to write dU px,yq in place of dCU pπU pxq,πU pyqq, and similarly for subsets of X. Moreover, we can always assume that X and the associated hyperbolic spaces are graphs (for example by [CdlH16, Lemma 3.B.6]). In particular, we can and shall assume that X and the CU are geodesic. We say that X admits an HHS structure if there is an HHS whose underlying metric space is X, and we write pX, Sq as shorthand for the entirety of a choice of HHS structure. An HHG is a finitely generated group G whose Cayley graph admits an HHS structure pG, Sq such that G acts cofinitely on S and elements of G induce isometries CU Ñ CgU for all U P S. (There are a couple of other natural regulatory assumptions that we shall not concern ourselves with here.) The idea behind two domains being orthogonal is that one can see a direct product of associated sub-HHSs inside X. This is made precise by the partial realisation axiom.

Axiom (Partial realisation). If tUiu is a set of pairwise orthogonal domains, then for any choice of points pi P CUi, there is some x P X with dUi px,piqď E for all i, and with Ui dV px, ρV qď E whenever Ui Ĺ V or Ui&V . In fact, one of the main tools for dealing with HHSs is the realisation theorem [BHS19, Theorem 3.1], which extends the partial realisation axiom. Roughly, it says that any consistent tuple is well-approximated by the projections of some point in X. In other words, performing constructions in X can be reduced to performing constructions in the associated hyperbolic spaces and checking that the points produced by this process are consistent.

Definition 3.1 (Consistent tuple). For a constant κ ě E, a tuple pbU q P UPS CU is said to be κ–consistent if V U ś min dU pbU , ρU q, dV pbV , ρV q ď κ whenever U&V, and U V min dV pbV , ρV q, diampbU Y ρU(pbV qq ď κ whenever U Ĺ V. 20 ( Axiom (Consistency). For any x P X, the tuple pπU pxqqUPS is E–consistent. It will be useful to be able to talk about consistency for subsets of S. Given u P CU and v P CV , we say that u and v satisfy the consistency inequalities for U and V if V U ‚ U&V and min dU pu, ρU q, dV pv, ρV q ď E, or U V min d v, ρU , diam u ρV v E ‚ (after relabelling) Ĺ and (V p V q pt uY U p qq ď . Let us now state the realisation theorem, which will be the mechanism for our proof ( of Theorem 3.5. We shall only need the existence part. Theorem 3.2 (Realisation, [BHS19, Theorem 3.1]). For each κ ě E, there are numbers θepκq and θupκq such that, if pbU qUPS is a κ–consistent tuple, then there is some x P X with dU px, bU q ď θepκq for all domains U. Moreover, the set of such x has diameter at most θupκq. A key application of the realisation theorem is for the construction of a coarse median operation for HHSs. Given three points x,y,z in an HHS pX, Sq, let pmU qUPS be the tuple whose U–entry is the median of the triple πU pxq,πU pyq,πU pzq in the hyperbolic space CU. This tuple is consistent [BHS19, Theorem 7.3], so we define µpx,y,zq to be a point obtained by applying the realisation theorem to the tuple pmU q. (One also needs a proposition of Bowditch [Bow13, Proposition 10.1] to conclude that pX,µ,dq is a coarse median space.) When X is an HHG, one can arrange for µ to be equivariant. The action on the index set is what distinguishes HHGs from groups that are HHSs, and this turns out to be an important distinction. For example, the property of being an HHS is invariant under quasi-isometries, but there are groups that are virtually HHGs but not HHGs themselves. Indeed, the p3, 3, 3q triangle group is virtually abelian, but, as mentioned in the introduction, it is not coarsely injective [Hod20a], and it therefore cannot be an HHG by Corollary H. A more direct proof not relying on the results of this paper is given in [PS20]. On the other hand, any group that is an HHS can be equipped with a coarse median [BHS19], but this is may fail to be equivariant if the structure is only an HHS structure. A related notion that is closed under taking subgroups is that of a group that acts on an HHS pX, Sq by HHS automorphisms. In other words, it acts on X isometrically, and on S with the regulatory assumptions alluded to above, but the action on S need not be cofinite. The median is still equivariant for such actions. In the theory of hyperbolic spaces, an important class of subsets are the quasiconvex subsets, because they inherit the structure of the ambient space. The natural analogue in the setting of hierarchical hyperbolicity is that of a hierarchically quasiconvex subset. Definition 3.3 (Hierarchical quasiconvexity). A subset Y of an HHS pX, Sq is said to be hierarchically quasiconvex if there is a function k such that: every πU pY q is kp0q– quasiconvex; and if x P X has dU px,Y qď r for all U P S, then dX px,Y qď kprq. We finish this section with some examples. All hyperbolic groups are hierarchically hyperbolic, as are the (extended) mapping class groups of finite type surfaces [BHS17b]; Teichmüller space with either of the standard metrics [BHS17b]; many graphs defined from curves on surfaces, including the pants graph [Vok17]; quotients of mapping class groups by powers of pseudo-Anosovs [BHS17a] 21 and Dehn-twist subgroups [BHMS20]; extensions of Veech groups [DDLS20]; the genus- two handlebody group [Mil20]; fundamental groups of closed 3–manifolds without Nil or Sol components [BHS19]; right angled Artin groups [BHS17b]; and in fact all known cubical groups [HS20]. Aside from the extensions of Veech groups and some 3–manifold groups, the groups listed here are all known to be HHGs, not merely HHSs. There are also various ways to combine HHSs and HHGs to produce new ones. For example, both classes are closed under relative hyperbolicity [BHS19]; any graph product of HHGs is an HHG [BR20]; and many graphs of groups are HHGs [BHS19, BR18, RS20]. 3.2. Coarse injectivity. Here we prove our result on hierarchically quasiconvex subsets of an HHS and deduce that HHSs are coarsely injective when equipped with the metric σ from Section 2. We then deduce that every HHG acts properly cocompactly by isometries on a coarsely injective space. We shall make use of the following powerful result for hyperbolic spaces. The version stated here is a combination of [CDV17, Lemma 5.1] and the proof of [CDV17, Theo- rem 5.1]. It states in particular that quasiconvex subsets of a hyperbolic graph satisfy a coarse version of the Helly property. Throughout this section, we say that subsets Z1 and Z2 of a metric space pX, dq are r–close if there exist z1 P Z1 and z2 P Z2 with dpz1, z2qď r. Theorem 3.4 ([CDV17]). The following holds for any nonnegative constants E, r, and k0. Let Y be an E–hyperbolic graph and let y be a vertex of Y . Suppose that Q is a 0 collection of pairwise 2Er–close k0–quasiconvex subsets of Y p q with the property that tdpy,Qq : Q P Qu is bounded. By discreteness, we can fix Q P Q with dpy,Qq maximal. Let z P Q have dpy, zq“ dpy,Qq, and let c be the point on a geodesic ry, zs with dpc, zq“ mintEr, dpy, zqu. Then dpc, Q1qď r1 for all Q1 P Q, where r1 “ maxt2k0 ` 5E, Er ` k0 ` 3Eu. The strength of this theorem is twofold. Firstly, the constant r1 is independent of the size of the set Q—a statement with this independence does not seem to appear elsewhere in the geometric group theory literature. The second strength is that the construction of the point c is both completely explicit and allows for a lot of flexibility in the choice of y. Observe that the condition that tdpy,Qq : Q P Qu is bounded is satisfied automatically if any Q P Q is bounded. We will now prove that hierarchically quasiconvex subsets of a hierarchically hyperbolic space satisfy a coarse version of the Helly property. Theorem 3.5 (Coarse Helly property). Let pX, Sq be an HHS with constant E, and let Q be a collection of k–hierarchically quasiconvex subsets of X such that either Q is finite or Q contains an element with bounded diameter. Suppose that there is a constant r such that any two elements of Q are r–close. There is a constant R “ RpE, k, rq such that there is a point x P X with dpx,Qqď R for all Q P Q. Proof. Let us say that a domain U begets a domain V if either U&V or U Ĺ V . If U U begets V then there is a well-defined bounded set ρV . Let U “ tU1,...,Unu be a maximal collection of pairwise orthogonal, nest-minimal domains. Note that we may choose U arbitrarily. For any domain V P S r U, there 22 is some i such that Ui begets V . By [DHS17, Lemma 1.5], for any domain V P S Ui Uj we have dV pρV , ρV q ď 2E whenever Ui and Uj both beget V . Moreover, recall that Ui diam ρV ď E. At the cost of increasing the hierarchical hyperbolicity constant to at Ui most 10E, we can therefore perturb the HHS structure to assume that every ρV is a Ui Uj U singleton, and that ρV “ ρV whenever both Ui and Uj beget V . We write ρV for the singleton U Ui ρV “ ρV . i : U begets V t i ď u U S r As mentioned, the construction of U ensures that the point ρV exists for all V P U. We are free to assume that if r ą 0 then r ą 1. Thus, by definition of hierarchical quasiconvexity and the fact that projection maps are pE, Eq–coarsely Lipschitz, we have that, for any domain V , the sets πV pQq, for Q P Q, are pairwise 2Er–close and k0– quasiconvex, where k0 “ kp0q. We assumed that either Q is finite or it contains an element with bounded diameter, so for any point y P X and any domain V , the set 1 tdV py,Qq : Q P Qu is bounded. Let r be as in the statement of Theorem 3.4. That 1 theorem now allows us to choose, for each U P U, a point bU in CU with dU pbU ,Qqď r for all Q P Q. For any other domain V , let bV be the point of CV obtained by applying Theorem 3.4 in the hyperbolic graph CV , with quasiconvex subsets tπV pQq : Q P Qu and U starting vertex ρV . 1 Claim: The tuple pbV qV PS is pr ` 7E ` Erq–consistent. W U Proof of Claim: Suppose that W begets V , and that dV pρV , ρV q ď 2E. Assume W 1 that dV pbV , ρV qą r `7E Èr. By the construction of bV , there exists some Q P Q such W W that dV pbV ,Qqď Er. As a consequence, we have dV pQ, ρV qě dV pbV , ρV q´dV pbV ,Qq´ W 1 V diam ρV ą r ` 6E. If W &V , then πW pQq is contained in the E–neighbourhood of ρW V 1 1 by consistency for elements of Q. In particular, dW pρW , bW qď r È as bW is r –close to 1 πW pQq. If W Ĺ V , then since πV pQq is k0–quasiconvex and r ` 6E ą k0 ` 6E, bounded V geodesic image and consistency show that the set ρW pπV pQqq has diameter at most E, and its E–neighbourhood contains πW pQq. Moreover, its E–neighbourhood contains V ρW pbV q by bounded geodesic image, as witnessed by the geodesic used to construct bV . Thus V V V diampbW Y ρW pbV qq ď dW pbW ,Qq` diam πW pQq` dW pQ, ρW pbV qq ` diam ρW pbV q ď r1 ` 3E ` 3E ` E “ r1 ` 7E. The above paragraph will be referred to as p˚q for the rest of the proof of the claim. We split the checking of the consistency inequalities for pairs pV,W q of domains into three cases. Case 1. W P U begets V . W U In this case, ρV “ ρV , so we are done by p˚q. Case 2. There is some U P U that begets both V and W . U U Proposition 1.8 of [BHS19] states that if W begets V then ρV and ρW satisfy the consistency inequalities for V and W . Consequently, by p˚q, the only case we need to 23 U V U check here is when U&W , W Ĺ V , and diampρW Y ρW pρV qq ď 2E. Assuming that W 1 U dV pρV , bV qą r ` 7E ` Er, there are two possibilities, depending on the location of ρV . U W If there is a geodesic rρV , bV s that is disjoint from the E–neighbourhood of ρV , then V U V U V diampρW pρV qYρW pbV qq ď E, so dW pρW , ρW pbV qq ď 3E. Moreover, for each Q P Q there is some q P Q such that any geodesic rbV ,πV pqqs is disjoint from the E–neighbourhood W V U of ρV . In particular, ρW pbV q is 2E–close to each πW pqq, and hence ρW is 5E–close to U each πW pQq. Since bW lies on a shortest geodesic between ρW and some πW pQq, we get U V that dW pbW , ρW qď 5E, and so bW is 8E–close to ρW pbV q. U W Otherwise, every geodesic rρV , bV s meets the E–neighbourhood of ρV . By construction W 1 1 of bV , there exists Q P Q such that dV pρV ,Qq ą pr `7EÈrqÈr´2E “ r `5E`2Er. V By the same argument as in p˚q, we now get that ρW pbV q is 3E–close to πW pQq, which V 1 has diameter at most 3E. Hence diampbW Y ρW pbV qq ď r ` 7E.

Case 3. No Ui begets both V and W , and neither V nor W is in U. After relabelling we can assume that U1 begets V and U2 begets W . Since U1 does not beget W we have U1KW , and similarly U2KV . In particular, the only case that needs checking is when V &W . The partial realisation axiom applied to any points p1 P CU1 U1 U2 and p2 P CU2 provides a point z P X such that dV pz, ρV qď E and dW pz, ρW qď E. By W U1 V U2 consistency for z, we have that either dV pρV , ρV qď 2E or dW pρW , ρW qď 2E. We are done by p˚q. ♦

In light of the claim, Theorem 3.2 provides a point x P X such that dV px, bV q ď 1 θepr `7E `Erq for all V P S. By construction of the points bV , we have that dV px,Qqď 1 1 r ` θepr ` 7E ` Erq for all Q P Q. Hierarchical quasiconvexity of the Q now tells us 1 1 that x is kpr ` θepr ` 7E ` Erqq–close to Q for all Q P Q.

It is worth noting that the proof of Theorem 3.5 gives ﬂexibility of a similar kind to that in Theorem 3.4. Indeed, we are free in our choice of U, and once this is chosen we apply the Chepoi–Dragan–Vaxès construction in each of the hyperbolic spaces associated with U, without restriction on the choice of starting point therein. We shall not need to make use of this in the present paper.

Corollary 3.6. If X is an HHS, then pX,σq is coarsely injective, hence roughly geodesic.

Proof. By Proposition 2.12, the geodesic coarse median space pX,µ,dq has quasicubical intervals, so Theorem 2.13 tells us that the metric σ is weakly roughly geodesic on X, that it is quasi-isometric to d, and that σ–balls are uniformly coarsely median–convex. Let tBσpxi, riq : i P Iu be a family of balls in pX,σq with the property that σpxi,xjqď ri `rj for all i, j P I. Since σ is weakly roughly geodesic, there is a constant δ, independent of the family of balls, such that the balls Bσpxi, ri ` δq intersect pairwise. Let Bi be the image of the ball Bσpxi, ri ` δq under the identity quasi-isometry pX,σq Ñ pX, dq. The Bi are uniformly coarsely median-convex, and so they are uniformly hierarchically quasiconvex by [RST18, Proposition 5.11]. They also intersect pairwise, and each is bounded, so Theorem 3.5 produces a point at uniformly bounded d–distance from each Bi. As d and σ are quasi-isometric, this point is at uniformly bounded σ–distance from each Bσpxi, ri ` δq. Thus pX,σq is coarsely injective. 24 Since any injective space is geodesic, we deduce that the coarsely injective metric space pX,σq is not merely weakly roughly geodesic, but actually roughly geodesic, as it is coarsely dense in its injective hull. Usually it really is necessary to change the metric: [CCG`20, Ex. 5.13] shows that Z3 with the standard ℓ1 metric is not coarsely injective, though it is an HHG. We now explain how to deduce the existence of a bicombing from work of Lang. See Section 1.3 for the definitions of roughly conical and roughly reversible bicombings. Corollary 3.7. If pX, Sq is an HHS, then pX,σq admits a roughly conical, roughly reversible, bicombing by rough geodesics that is coarsely equivariant under the automorphism group of pX, Sq. Proof. According to Corollary 3.6, the metric space pX,σq is coarsely injective, so it is D–coarsely dense in its injective hull for some D. A construction of Lang shows that every injective metric space E admits a conical, reversible, geodesic, Isom E–invariant 1 bicombing γ [Lan13]. Take E “ EppX,σqq. For each a, b P X and t P r0, 1s, define γa,bptq 1 as any point of X at distance at most D from γa,bptq. Since γptq is at uniform distance D from γ1ptq, we deduce that γ is a bicombing on pX,σq with the listed properties. Note that if the action of the automorphism group of pX, Sq on X is free, then the bicombing may be chosen to be actually equivariant. Let us now discuss the consequences of our construction for HHGs. Corollary 3.8. If G is an HHG, then G admits a proper, cocompact, isometric action on the coarsely injective space pG, σq. Proof. pG, σq is coarsely injective by Corollary 3.6. Since the median is equivariant in an HHG, Lemma 2.10 tells us that the action is isometric. Properness and cocompactness follow from Proposition 2.16. Remark 3.9. In fact, we do not quite need to assume that we have a hierarchically hyperbolic group in Corollary 3.8: we only need a proper cocompact action by median isometries on an HHS. In fact, cocompactness can be relaxed to coboundedness for the sake of the applications in this paper. For example, it would be sufficient to assume that G is a group acting properly coboundedly by HHS automorphisms on an HHS. The consequences for HHGs listed here and in the introduction therefore apply in this generality. The next lemma is a modified version of [CCG`20, Proposition 6.7], in which the assumption that the hull is proper has been dropped. Lemma 3.10. If a group G acts properly coboundedly on a coarsely injective space X, then G acts properly coboundedly on the injective hull EpXq. In particular, every HHG admits a proper, cobounded action on an injective space. Proof. There is an induced action of G on EpXq and the isometric embedding e: X Ñ EpXq is equivariant with respect to this induced action [Lan13, Proposition 3.7]. To simplify notation, we identify the points of X with their images under e and thus identify X with EpXq. The Hausdorff distance between X and EpXq is bounded by some constant 25 D, so the action of G on EpXq is cobounded. For properness, let Y Ă EpXq be bounded and let Y 1 “ tx P X : dpY,xq ď Du ‰ ∅. Since e is an isometric embedding, Y 1 is bounded. If g P G has gY X Y ‰ ∅, then pick y P Y with gy P Y and let x P X have dpy,xqď D. Then dpgy,gxqď D, so gx is D–close to Y . That is, gY 1 X Y 1 ‰ ∅, so since Y 1 is bounded and the action of G on X is proper, there are only finitely many such g. The final sentence follows from Corollary 3.8. Next we strengthen Corollary 3.7 in the case of HHGs. In particular, this applies to (extended) mapping class groups of finite type surfaces. Corollary 3.11. If G is an HHG, then G is semihyperbolic. Proof. By Lemma 3.10, G acts properly coboundedly on an injective space E. Every orbit map G Ñ E is a G–equivariant quasi-isometry. By [Lan13, Proposition 3.8], E has a G–invariant, bounded, geodesic bicombing in the sense of [AB95]. As the action of G on itself is free, it is semihyperbolic by [AB95, Theorem 4.1]. Corollary 3.12. If G is an HHG, then G has finitely many conjugacy classes of finite subgroups. Proof. By Lemma 3.10, G acts properly coboundedly on an injective space E. Let x P E and let r be a constant such that G¨x is r–coarsely dense in E. Let F be a finite subgroup of G. By [Lan13, Proposition 1.2], there is a point z P E that is fixed by F , and hence F fixes the ball Bpz, rq in E, which contains a point of G ¨ x. It follows that a conjugate of F fixes a point in Bpx, rq, and we are done by properness of the action. 3.3. Packing subgroups. Here we describe the application to bounded packing mentioned in the introduction. Following Hruska and Wise [HW09], we say that a finite collection H of subgroups of a discrete group G has bounded packing in G if for each N there is a constant r such that for any collection of N distinct cosets of elements of H, at least two are separated by a distance of at least r (with respect to some left-invariant, proper distance). If H consists of a single subgroup H, then we say that H has bounded packing in G. Corollary 3.13. If H is a finite collection of hierarchically quasiconvex subgroups of a group G that is an HHS, then H has bounded packing in G. Proof. By Theorem 3.5, any finite collection of cosets of elements of H that are pairwise r–close must all come R–close to a single point x P G. In other words, they all intersect the R–ball about x. Since distinct cosets of a given subgroup are disjoint and balls in G are finite, this bounds the size of the collection of cosets. In the case of quasiconvex subgroups of hyperbolic groups, one can use Theorem 3.4 in place of Theorem 3.5 in this argument to provide a new, simpler proof of bounded packing. This type of argument is also implicit in [HP19, Remark 4.4, Corollary 4.5], though the coarse Helly property for quasiconvex subgroups of hyperbolic groups is established in a much less efficient way there. Previous proofs of this result work by induction on the height of subgroups. However, this line of reasoning does not generalise outside the setting of strict negative curvature; indeed, no subgroup of a flat can ever have finite height. Moreover, Theorems 3.4 and 3.5 26 are purely geometric: there is no group action involved. It therefore seems that the most natural way to establish bounded packing for quasiconvex subgroups of hyperbolic groups is via the Chepoi–Dragan–Vaxès theorem as described above. If a group G has a codimension–1 subgroup H, then Sageev’s construction yields an action of G on a CAT(0) cube complex, and if the conjugates of H satisfy the coarse Helly property, then it follows that the action of G on the CAT(0) cube complex is cocompact ([Sag97]). This raises the following question.

Question. Does the mapping class group have property FW8, i.e. does any action of the mapping class group on a ﬁnite-dimensional CAT(0) cube complex have a ﬁxed point?

Note that property FW8 is intermediate between having no virtual surjection onto Z and Kazhdan’s property (T). There are known restrictions on what an action of the mapping class group on a CAT(0) cube complex could look like. Indeed, the mapping class group of a surface of genus at least three does not admit a properly discontinuous action by semisimple isometries on a complete CAT(0) space ([KL96, BH99, Bri10]), nor, more speciﬁcally, does it act properly on a CAT(0) cube complex (even an inﬁnite dimensional one) [Gen19]. More generally, in relationship with property (T) and the Haagerup property, the existence of non-trivial actions of the mapping class group on various generalisations of CAT(0) cube complexes remains mysterious; for example median spaces, Hilbert spaces, CAT(0) spaces, and Lp spaces. The coarse version of the Helly property established here may prove useful in the study of such actions.

4. Strong Shortcut Property In this section we will prove that coarsely injective spaces of uniformly bounded geometry are strongly shortcut. Recall that a metric space has uniformly bounded geometry if, for any r ą 0, there exists a uniform Nprq P N such that every ball of radius r contains at most Nprq points. A Riemannian circle S is S1 endowed with a geodesic metric of some length |S|. A roughly geodesic metric space pX,σq is strongly shortcut if there exists K ą 1 such that for any C ą 0 there is a bound on the lengths |S| of pK,Cq–quasi-isometric embeddings S Ñ X of Riemannian circles S in pX,σq [Hod20b]. A group G is strongly shortcut if it acts properly and coboundedly on a strongly shortcut metric space [Hod18, Hod20b]. We will now give a brief description of the injective hull construction of Isbell [Isb64], which was later rediscovered by Dress [Dre84] and Chrobak and Larmore [CL94]. For a nice discussion on this construction, see Lang [Lan13]. Let pX,σq be a metric space. A radius function on X is a function f : X Ñ Rě0 for which σpx,yqď fpxq` fpyq for every x,y P X. A radius function f : A Ñ Rě0 on any subspace of A Ď X is called a partial radius function on X. If f, g : X Ñ Rě0 are two radius functions then f dominates g if fpxq ě gpxq for all x P X. A radius function f : X Ñ Rě0 is minimal if the only radius function it dominates is itself. If f : A Ñ Rě0 is a partial radius function on X then there exists a minimal radius function g : X Ñ Rě0 such that g|A is dominated by f. For any x P X, the function 27 σp¨,xq is a minimal radius function. If f, g : X Ñ Rě0 are two minimal radius functions then |f ´ g|8 “ sup fpxq´ gpxq xPX ˇ ˇ is ﬁnite. The set of minimal radius functionsˇ on X, withˇ metric given by dEpXqpf, gq “ |f ´ g|8, is the injective hull EpXq of X. The isometric embedding e: X ãÑ EpXq sends x P X to the minimal radius function epxq: y ÞÑ σpx,yq and for any x P X and f P EpXq we have dEpXq epxq,f “ fpxq.

Lemma 4.1. `Let pX,σ˘ q be a metric space. Let g : X Ñ Rě0 be a minimal radius ¯ function, let f : X Ñ Rě0 be a radius function and let f : X Ñ Rě0 be any minimal ¯ ¯ radius function dominated by f. Then |g ´ f|8 ď |g ´ f|8. ¯ ¯ ¯ Proof. Let y P X. Then fpyq ď fpyq ď gpyq ` |g ´ f|8 and so fpyq´ gpyq ď |g ´ f|8. ¯ It remains to prove that gpyq´ fpyq ď |g ´ f|8. By minimality of g, for any ǫ ą 0, there exists z P X for which gpyq` gpzqă σpy, zq` ǫ. Then, since f is a radius function dominated by f¯, we have fpyqě σpy, zq´ fpzq ě σpy, zq´ f¯pzq ¯ ě σpy, zq´ gpzq ´ |g ´ f|8 ¯ ą gpyq´ ǫ ´ |g ´ f|8 ¯ and so gpyq ´ fpyq ă |g ´ f|8 ` ǫ which completes the proof since we chose ǫ ą 0 arbitrarily. Theorem 4.2. Let pX,σq be a coarsely injective metric space. If pX,σq has uniformly bounded geometry then pX,σq is strongly shortcut. Proof. In order to prove this theorem, we will show that for some uniform radius r, a pK,Cq-quasi-isometric embedding of a Riemannian circle S Ñ X implies the existence of a “center” point x such that the cardinality of the ball Bpx, rq is bounded below by an expression that tends to infinity as K approaches 1 and |S| approaches infinity. If X is not strongly shortcut then, for any K ą 1, it will admit pK,CK q–quasi-isometric embeddings of arbitrarily long Riemannian circles so that we would then contradict the uniformly bounded geometry assumption. Let X Ñ EpXq be the embedding of pX,σq into its injective hull and view this embedding as an inclusion of a subspace. By Proposition 1.1, the subspace X is δ– coarsely dense in EpXq for some δ ą 0. So there is a retraction r : EpXqÑ X such that r is a p1, 2δq–quasi-isometry. Let φ: S Ñ X be a pK,Cq–quasi-isometric embedding of a Riemannian circle. Let 2 |S| 2 f : φpSq Ñ Rě0 be the constant function taking the value K 4 ` C. Then f is a 1 radius function on φpSqĂ X. Let f : φpSqÑ Rě0 be a minimal radius function on φpSq dominated by f 2. Then for each x P φpSq and each ǫ ą 0, there exists a y P φpSq for which f 1pxq` f 1pyq ă σpx,yq` ǫ. Since f 1 is a partial radius function on X we can let 1 f : X Ñ Rě0 be a minimal radius function on X dominated by f . Then f is a point of 1 EpXq, by definition of EpXq. Moreover, if x P φpSq then dEpXq f,xq“ fpxqď f pxq ď 28 ` 2 |S| ´1 f pxq“ K 4 ` C so if s¯ is the antipode in S of any element of s P φ pxq then we have 1 |S| d x, φps¯q “ d φpsq, φps¯q ě d ps, s¯q´ C “ ´ C EpXq EpXq K S 2K and ` ˘ ` ˘ |S| d x, φps¯q ď d x,f ` d f, φps¯q “ fpxq` f φps¯q ď fpxq` K ` C EpXq EpXq EpXq 4 ` ˘|S| |S`| ˘ 2´K2` ˘ ` ˘ so that fpxqě 2K ´ K 4 ´ 2C “ 4K |S|´ 2C. Thus we have shown that 2 ´ K2 |S| |S|´ 2C ď fpxqď K ` C 4K 4 for any x P φpSq. 1 1 For x,y P X let ℓx,y “ fpxq` fpyq´ σpx,yq. Since f is dominated by f and f is a minimal radius function on φpSq, for each x P φpSq and each ǫ ą 0, there exists y P φpSq such that ℓx,y ă ǫ. Moreover, for a, b P S we have 2 ´ K2 |S|´ 4C ď f φpaq ` f φpbq 2K “ σ`φpaq˘, φpbq` ` ℓφ˘paq,φpbq ď Kd` Spa, bq`˘C ` ℓφpaq,φpbq 2 5 2´K ℓφpaq,φpbq` C and so dSpa, bqě 2K2 |S|´ K . k Claim: Let x P φpSq. There exists a sequence of minimal radius functions pfx : X Ñ R fpxq ě0qk where k ranges in t0, 1,...,Mxu such that Mx “ δ and the following properties hold for all k, k1 and y. 0 X \ (1) fx “ f k k1 1 (2) dEpXqpfx ,fx q“ δ|k ´ k | k (3) fpyq` kδ ´ ℓx,y ď fx pyqď fpyq` maxt0, kδ ´ ℓx,yu k Proof of Claim: We construct the pfx qk by induction on k. By Property (1), we 0 k´1 must start with fx “ f. Assuming we have fx , we will begin by defining a radius ¯k ¯k k´1 k´1 function fx . Set fx pxq “ fx pxq´ δ. By minimality of fx , there exists y P X for which the inequality k´1 k´1 (2) fx pyq` fx pxq´ δ ă σpx,yq holds. Indeed, if no such y existed then k´1 fx pyq if y ‰ x y ÞÑ k´1 δ #fx pyq´ 2 if y “ x k´1 would be a radius function that is dominated by but not equal to fx and this would k´1 ¯k k´1 contradict minimality of fx . Set fx pxq “ fx pxq´ δ. For all y P Xztxu satisfying ¯k k´1 ¯k Equation (2), set fx pyq “ σpx,yq´ fx pxq` δ. For all other y P Xztxu, set fx pyq “ k´1 ¯k k´1 ¯k fx pyq. Then, except for at y “ x, we have fx pyq ě fx pyq. Thus, to check that fx ¯k ¯k is a radius function, we need only verify that fx pxq` fx pyq ě σpx,yq for any y P X. ¯k ¯k k´1 When y “ x the inequality fx pxq ` fx pyq ě σpx,yq is equivalent to fx pxq ě δ, 29 which holds by the inductive application of Property (2) and the triangle inequality. ¯k ¯k When y satisfies Equation (2) the inequality fx pxq` fx pyq ě σpx,yq is equivalent to k´1 k´1 fx pxq´ δ ` σpx,yq´ fx pxq` δ ě σpx,yq, which holds with equality. Finally, when ¯k ¯k k´1 k´1 y does not satisfy Equation (2), we have fx pxq ` fx pyq “ fx pxq´ δ ` fx pyq ě k´1 k´1 ¯k k fx pxq´ δ ` σpx,yq´ fx pxq` δ “ σpx,yq. Thus fx is a radius function. Define fx ¯k as any minimal radius function that is dominated by fx . ¯k k´1 ¯k Since fx pyq“ σpx,yq´ fx pxq` δ “ σpx,yq´ fx pxq for some y P X, we must have k ¯k k´1 k´1 k fx pxq“ fx pxq“ fx pxq´ δ. Thus |fx ´ fx |8 ě δ and fpxq d pf Mx ,xq“ f Mx pxq“ fpxq´ M δ “ fpxq´ δ ă δ EpXq x x x δ Y ] Mx k´1 k so we have dEpXqpfx ,xqă δ. On the other hand, by Lemma 4.1, we have |fx ´fx |8 ď k´1 ¯k k´1 k k´1 k |fx ´ fx |8 ď δ and so dEpXqpfx ,fx q “ |fx ´ fx |8 “ δ. Therefore,

dEpXqpf,xq“ fpxq “ Mxδ ` fpxq´ Mxδ

Mx “ Mxδ ` dEpXqpfx ,xq

Mx k´1 k Mx “ dEpXqpfx ,fx q` dEpXqpfx ,xq k“1 ÿ 0 where fx “ f. Then, by the triangle inequality, Property (2) is satisﬁed. To verify Property (3), let y P X. We have

k k fpyq` kδ ´ ℓx,y “ σpx,yq` kδ ´ fpxq“ σpx,yq´ fx pxqď fx pyq

k so the lower bound holds. The upper bound on fx pyq given by Property (3) is Rk “ fpyq ` maxt0, kδ ´ ℓx,yu. Suppose Property (3) doesn’t hold and let k be the least k k k´1 integer for which fx pyq ą Rk. Then k ą 0 and k must satisfy fx pyq ´ fx pyq ą k k k´1 Rk ´ Rk´1 ě 0. By the construction of fx , the fact that fx pyq ą fx pyq implies that k´1 k´1 ¯k k´1 ¯k fx pyq`fx pxq´δ ă σpx,yq and that fx pyq“ σpx,yq´fx pxq`δ “ σpx,yq´fx pxq. Then we must have

k ¯k fx pyq“ fx pyq ¯k “ σpx,yq´ fx pxq k “ σpx,yq´ fx pxq “ fpyq` kδ ´ ℓx,y

ď Rk

k which contradicts fx pyqą Rk. Thus we have verified Property (3). ♦ k We will now use the sequence pfx qk of minimal radius functions to prove the theorem. 2 1 1 2pK ´1q 4δ`10C 1 Assume that a, a P S satisfy dSpa, a qě K2 |S|` K . Such a and a exist when 1 1 K is close enough to 1. Take b P S for which ℓφpaq,φpbq ă δ. Then dSpa, a q` dSpa , bq` 30 dSpb, aq ď |S| so we have 1 1 dSpa , bq ď |S|´ dSpa, bq´ dSpa, a q 2 2 ´ K ℓφpaq,φpbq ` 5C ď |S|´ |S|` ´ d pa, a1q 2K2 K S 2 ´ K2 5C ` δ ă |S|´ |S|` ´ d pa, a1q 2K2 K S 2 ´ K2 5C ` δ 2pK2 ´ 1q 4δ ` 10C ď |S|´ |S|` ´ |S|´ 2K2 K K2 K 2 ´ K2 3δ ` 5C “ |S|´ 2K2 K and so 2 2 2 ´ K ℓφpa1q,φpbq ` 5C 2 ´ K 3δ ` 5C |S|´ ď d pa1, bqă |S|´ 2K2 K S 2K2 K which implies ℓφpa1q,φpbq ą 3δ. So we have 3 fφpa1q φpbq ď f φpbq ` max 0, 3δ ´ ℓφpa1q,φpbq ` ˘ “ f`φpbq˘ ( 3 ď fφ`paq φ˘pbq ´ 3δ ` ℓφpaq,φpbq 3 ă fφpaq`φpbq˘ ´ 2δ where the inequalities are applications of` property˘ (3). Thus 3 3 dEpXqpfφpa1q,fφpaqqą 2δ 3 3 and so rpfφpa1qq and rpfφpaqq are distinct elements of the metric ball B rpfq, 5δ of radius N 5δ centered at rpfq in X. So, if taiui“1 are points of S that subdivide S into segments of 2 ` ˘ 2pK ´1q 4δ`10C length at least K2 |S|` K then B rpfq, 5δ contains at least N points. Subdi- 2 ´1 2pK ´1q 4δ`10C viding S evenly we can achieve N “ K`2 ` K˘|S| . So we have shown that if X admits a pK,Cq–quasi-isometric embeddingY´ of a Riemannian¯ ] circle S and K is close 2 ´1 2pK ´1q 4δ`10C enough to 1 then for some x P X we have Bpx, 5δq ě K2 ` K|S| . To complete the proof, suppose X is notˇ stronglyˇ Y´ shortcut. Then, for¯ each] K ą 1, there exists CK ą 0 and a sequence ˇpφn : Sn ˇÑ Xqn of pK,CK q–quasi-isometric embeddings of Riemannian circles where |Sn|ě n. The argument above shows that, for each small enough K ą 1 and each n P N there exists xK,n P X satisfying BpxK,n, 5δq ě 1 1 2 2 1 ´ 2 2 1 ´ 2 pK ´ q 4δ`10Ck pK ´ q 4δ`10CK K 2 ` . The expression 2 ` tends to 2 as K K|Sn| K K|Sn| ˇ 2pK ´1qˇ ˇ 2 ˇ Y´ ¯ ] N ´ ¯ K n tends to infinity so if nK P is large enough then BpxK,nK , 5δq ě 2pK2´1q ´ 1. But K2 this contradicts the uniform bounded geometry assumptionˇ on X sinceˇ 2 2 1 tends to ˇ ˇ pK ´ q infinity as K tends to 1. References

[AB95] Juan M. Alonso and Martin R. Bridson. Semihyperbolic groups. Proc. London Math. Soc. (3), 70(1):56–114, 1995. 31 [AB18] Carolyn Abbott and Jason Behrstock. Conjugator lengths in hierarchically hyperbolic groups. arXiv preprint arXiv:1808.09604, 2018. [ABD17] Carolyn Abbott, Jason Behrstock, and Matthew Gentry Durham. Largest acylindrical actions and stability in hierarchically hyperbolic groups. arXiv preprint arXiv:1705.06219, 2017. [AMST19] Yago Antolín, Mahan Mj, Alessandro Sisto, and Samuel J. Taylor. Intersection properties of stable subgroups and bounded cohomology. Indiana Univ. Math. J., 68(1):179–199, 2019. [ANS19] Carolyn Abbott, Thomas Ng, and Davide Spriano. Hierarchically hyperbolic groups and uniform exponential growth. arXiv preprint arXiv:1909.00439, 2019. [AP56] N. Aronszajn and P. Panitchpakdi. Extension of uniformly continuous transformations and hyperconvex metric spaces. Pacific J. Math., 6:405–439, 1956. [BD14] Jason Behrstock and Cornelia Druţu. Divergence, thick groups, and short conjugators. Illinois J. Math., 58(4):939–980, 2014. [BH99] Martin R. Bridson and André Haefliger. Metric spaces of non-positive curvature, volume 319 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 1999. [BHMS20] Jason Behrstock, Mark Hagen, Alexandre Martin, and Alessandro Sisto. A combinatorial take on hierarchical hyperbolicity and applications to quotients of mapping class groups. arXiv preprint arXiv:2005.00567, 2020. [BHS17a] Jason Behrstock, Mark Hagen, and Alessandro Sisto. Asymptotic dimension and small- cancellation for hierarchically hyperbolic spaces and groups. Proc. Lond. Math. Soc. (3), 114(5):890–926, 2017. [BHS17b] Jason Behrstock, Mark Hagen, and Alessandro Sisto. Hierarchically hyperbolic spaces, I: Curve complexes for cubical groups. Geom. Topol., 21(3):1731–1804, 2017. [BHS17c] Jason Behrstock, Mark Hagen, and Alessandro Sisto. Quasiflats in hierarchically hyperbolic spaces. arXiv preprint arXiv:1704.04271, 2017. [BHS19] Jason Behrstock, Mark Hagen, and Alessandro Sisto. Hierarchically hyperbolic spaces II: Combination theorems and the distance formula. Pacific J. Math., 299(2):257–338, 2019. [BM11] Jason A. Behrstock and Yair N. Minsky. Centroids and the rapid decay property in mapping class groups. J. Lond. Math. Soc. (2), 84(3):765–784, 2011. [Bow13] Brian H. Bowditch. Coarse median spaces and groups. Pacific J. Math., 261(1):53–93, 2013. [Bow19a] Brian H. Bowditch. Convex hulls in coarse median spaces. http://www.warwick.ac.uk/~masgak/papers/hulls-cms.pdf, 2019. [Bow19b] Brian H. Bowditch. Quasiflats in coarse median spaces. http://www.warwick.ac.uk/~masgak/papers/quasiflats.pdf, 2019. [Bow20] Brian H. Bowditch. Median and injective metric spaces. Math. Proc. Cambridge Philos. Soc., 168(1):43–55, 2020. [BR18] Federico Berlai and Bruno Robbio. A refined combination theorem for hierarchically hyperbolic groups. arXiv preprint arXiv:1810.06476, 2018. [BR20] Daniel Berlyne and Jacob Russell. Hierarchical hyperbolicity of graph products. arXiv preprint arXiv:2006.03085, 2020. [Bri00] Martin R. Bridson. Finiteness properties for subgroups of GLpn, Zq. Math. Ann., 317(4):629– 633, 2000. [Bri10] Martin R. Bridson. Semisimple actions of mapping class groups on CATp0q spaces. In Geom- etry of Riemann surfaces, volume 368 of London Math. Soc. Lecture Note Ser., pages 1–14. Cambridge Univ. Press, Cambridge, 2010. [Bri19] Martin R. Bridson. Semihyperbolicity. In Beyond hyperbolicity, volume 454 of London Math. Soc. Lecture Note Ser., pages 25–64. Cambridge Univ. Press, Cambridge, 2019. [BvdV91] H.-J. Bandelt and M. van de Vel. Superextensions and the depth of median graphs. J. Combin. Theory Ser. A, 57(2):187–202, 1991. [CCG`20] Jérémie Chalopin, Victor Chepoi, Anthony Genevois, Hiroshi Hirai, and Damian Osajda. Helly groups. arXiv preprint arXiv:2002.06895, 2020. [CdlH16] Yves Cornulier and Pierre de la Harpe. Metric geometry of locally compact groups, volume 25 of EMS Tracts in Mathematics. European Mathematical Society (EMS), Zürich, 2016.

32 [CDV17] Victor Chepoi, Feodor F. Dragan, and Yann Vaxès. Core congestion is inherent in hyperbolic networks. In Proceedings of the Twenty-Eighth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2264–2279. SIAM, Philadelphia, PA, 2017. [CE07] Victor Chepoi and Bertrand Estellon. Packing and covering δ-hyperbolic spaces by balls. In Approximation, Randomization, and Combinatorial Optimization. Algorithms and Tech- niques, pages 59–73. Springer, 2007. [Che94] Victor Chepoi. Separation of two convex sets in convexity structures. J. Geom., 50(1-2):30–51, 1994. [Che00] Victor Chepoi. Graphs of some CATp0q complexes. Adv. in Appl. Math., 24(2):125–179, 2000. [CL94] Marek Chrobak and Lawrence L. Larmore. Generosity helps or an 11-competitive algorithm for three servers. J. Algorithms, 16(2):234–263, 1994. [CS11] Pierre-Emmanuel Caprace and Michah Sageev. Rank rigidity for CAT(0) cube complexes. Geom. Funct. Anal., 21(4):851–891, 2011. [DDLS20] Spencer Dowdall, Matthew G. Durham, Christopher J. Leininger, and Alessandro Sisto. Extensions of Veech groups are hierarchically hyperbolic. arXiv preprint arXiv:2006.16425, 2020. [DHS17] Matthew Gentry Durham, Mark F. Hagen, and Alessandro Sisto. Boundaries and automorphisms of hierarchically hyperbolic spaces. Geom. Topol., 21(6):3659–3758, 2017. [DL15] Dominic Descombes and Urs Lang. Convex geodesic bicombings and hyperbolicity. Geom. Dedicata, 177:367–384, 2015. [DMS20] Matthew G. Durham, Yair N. Minsky, and Alessandro Sisto. Stable cubulations, bicombings and barycenters. arXiv preprint arXiv:2009.13647, 2020. [Dre84] Andreas W. M. Dress. Trees, tight extensions of metric spaces, and the cohomological dimension of certain groups: a note on combinatorial properties of metric spaces. Adv. in Math., 53(3):321–402, 1984. [DT15] Matthew Gentry Durham and Samuel J. Taylor. Convex cocompactness and stability in mapping class groups. Algebr. Geom. Topol., 15(5):2839–2859, 2015. [ECH`92] David B. A. Epstein, James W. Cannon, Derek F. Holt, Silvio V. F. Levy, Michael S. Paterson, and William P. Thurston. Word processing in groups. Jones and Bartlett Publishers, Boston, MA, 1992. [EW17] Alexander Engel and Christopher Wulﬀ. Coronas for properly combable spaces. arXiv preprint arXiv:1711.06836, 2017. [FO20] Tomohiro Fukaya and Shin-ichi Oguni. A coarse Cartan-Hadamard theorem with application to the coarse Baum-Connes conjecture. J. Topol. Anal., 12(3):857–895, 2020. [Gen19] Anthony Genevois. A cubical ﬂat torus theorem and some of its applications. arXiv preprint arXiv:1902.04883, 2019. [Gen20] Anthony Genevois. Hyperbolicities in CAT(0) cube complexes. Enseign. Math., 65(1-2):33– 100, 2020. [GMRS98] Rita Gitik, Mahan Mitra, Eliyahu Rips, and Michah Sageev. Widths of subgroups. Trans. Amer. Math. Soc., 350(1):321–329, 1998. [Hae20] Thomas Haettel. Hyperbolic rigidity of higher rank lattices. Ann. Sci. Éc. Norm. Supér. (4), 53(2):439–468, 2020. [Ham09] Ursula Hamenstädt. Geometry of the mapping class group II: A biautomatic structure. arXiv preprint arXiv:0912.0137, 2009. [HO19] Jingyin Huang and Damian Osajda. Helly meets garside and artin. arXiv preprint arXiv:1904.09060, 2019. [Hod18] Nima Hoda. Shortcut graphs and groups. arXiv preprint arXiv:1811.05036, 2018. [Hod20a] Nima Hoda. Crystallographic Helly groups. arXiv preprint arXiv:2010.07407, 2020. [Hod20b] Nima Hoda. Strongly shortcut spaces. arXiv preprint arXiv:2010.07400, 2020. [HP19] Mark Hagen and Harry Petyt. Projection complexes and quasimedian maps. people.maths.bris.ac.uk/~aj18755/hierarchically-quasitree.pdf, 2019. [HS20] Mark F. Hagen and Tim Susse. On hierarchical hyperbolicity of cubical groups. Israel J. Math., 236(1):45–89, 2020.

33 [HW09] G. Christopher Hruska and Daniel T. Wise. Packing subgroups in relatively hyperbolic groups. Geom. Topol., 13(4):1945–1988, 2009. [HW14] G. Christopher Hruska and Daniel T. Wise. Finiteness properties of cubulated groups. Com- pos. Math., 150(3):453–506, 2014. [Isb64] J. R. Isbell. Six theorems about injective metric spaces. Comment. Math. Helv., 39:65–76, 1964. [Ker83] Steven P. Kerckhoﬀ. The Nielsen realization problem. Ann. of Math. (2), 117(2):235–265, 1983. [KL96] Michael Kapovich and Bernhard Leeb. Actions of discrete groups on nonpositively curved spaces. Math. Ann., 306(2):341–352, 1996. [Lan13] Urs Lang. Injective hulls of certain discrete metric spaces and groups. J. Topol. Anal., 5(3):297–331, 2013. [Mil20] Marissa Miller. Stable subgroups of the genus two handlebody group. arXiv preprint arXiv:2009.05067, 2020. [MM00] H. A. Masur and Y. N. Minsky. Geometry of the complex of curves. II. Hierarchical structure. Geom. Funct. Anal., 10(4):902–974, 2000. [Mos95] Lee Mosher. Mapping class groups are automatic. Ann. of Math. (2), 142(2):303–384, 1995. [NR98] G. A. Niblo and L. D. Reeves. The geometry of cube complexes and the complexity of their fundamental groups. Topology, 37(3):621–633, 1998. [NR03] Graham A. Niblo and Lawrence Reeves. Coxeter groups act on CATp0q cube complexes. J. Group Theory, 6(3):399–413, 2003. [NWZ19] Graham A. Niblo, Nick Wright, and Jiawen Zhang. A four point characterisation for coarse median spaces. Groups Geom. Dyn., 13(3):939–980, 2019. [PS20] Harry Petyt and Davide Spriano. Unbounded domains in hierarchically hyperbolic groups. arXiv preprint arXiv:2007.12535, 2020. [Rol98] Martin Roller. Poc sets, median algebras and group actions. Habilitationsschrift, Universität Regensburg, 1998. Available on the arXiv at arXiv:1607.07747. [RS99] Hyam Rubinstein and Michah Sageev. Intersection patterns of essential surfaces in 3- manifolds. Topology, 38(6):1281–1291, 1999. [RS20] Bruno Robbio and Davide Spriano. Hierarchical hyperbolicity of hyperbolic-2-decomposable groups. arXiv preprint arXiv:2007.13383, 2020. [RST18] Jacob Russell, Davide Spriano, and Hung Cong Tran. Convexity in hierarchically hyperbolic spaces. arXiv preprint arXiv:1809.09303, 2018. [Sag97] Michah Sageev. Codimension-1 subgroups and splittings of groups. J. Algebra, 189(2):377– 389, 1997. [Sis16] Alessandro Sisto. Quasi-convexity of hyperbolically embedded subgroups. Math. Z., 283(3- 4):649–658, 2016. [Tao13] Jing Tao. Linearly bounded conjugator property for mapping class groups. Geom. Funct. Anal., 23(1):415–466, 2013. [Tra19] Hung Cong Tran. On strongly quasiconvex subgroups. Geom. Topol., 23(3):1173–1235, 2019. [Vok17] Kate M Vokes. Hierarchical hyperbolicity of graphs of multicurves. arXiv preprint arXiv:1711.03080, 2017.

Institut Montpelliérain Alexander Grothendieck, CNRS, Univ. Montpellier., Place Eugène Bataillon, 34090 Montpellier, France Email address: [email protected]

Département de mathématiques et applications, École normale supérieure, 45 rue d’Ulm, 75005 Paris, France Email address: [email protected]

School of Mathematics, University of Bristol, UK Email address: [email protected]